aclSetDynamicTensorAddr

Function Usage

After aclOpExecutor reuse is enabled by calling aclSetAclOpExecutorRepeatable, if the input or output device memory address changes, the device memory address recorded in the corresponding aclTensorList needs to be updated.

Prototype

aclnnStatus aclSetDynamicTensorAddr(aclOpExecutor *executor, size_t irIndex, const size_t relativeIndex, aclTensorList *tensors, void *addr)

Parameters

Parameter	Input/Output	Description
executor	Input	aclOpExecutor that is set to the reusable state.
irIndex	Input	Index of aclTensorList to be updated in the operator prototype definition, starting from 0.
relativeIndex	Input	Index of aclTensor to be updated in aclTensorList. If aclTensorList has N tensors, the value range is [0, N – 1].
tensors	Input	aclTensorList pointer to be updated.
addr	Input	Device storage address to be updated to the specified aclTensor.

Returns

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

Possible causes:

If error code 561103 is returned, executor or tensors is a null pointer.
If error code 161002 is returned, the value of relativeIndex is greater than or equal to the number of tensors in the tensors list.
If error code 161002 is returned, the value of irIndex is greater than or equal to the number of input/output parameters for the operator prototype.

Constraints

None

Examples

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

// Create the input and output aclTensor and aclTensorList.
std::vector<int64_t> shape = {1, 2, 3};
aclTensor tensor1 = aclCreateTensor(shape.data(), shape.size(), aclDataType::ACL_FLOAT,
nullptr, 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), nullptr);
aclTensor tensor2 = aclCreateTensor(shape.data(), shape.size(), aclDataType::ACL_FLOAT,
nullptr, 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), nullptr);
aclTensor tensor3 = aclCreateTensor(shape.data(), shape.size(), aclDataType::ACL_FLOAT,
nullptr, 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), nullptr);
aclTensor tensor4 = aclCreateTensor(shape.data(), shape.size(), aclDataType::ACL_FLOAT,
nullptr, 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), nullptr);
aclTensor tensor5 = aclCreateTensor(shape.data(), shape.size(), aclDataType::ACL_FLOAT,
nullptr, 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), nullptr);
aclTensor *list1[] = {tensor1, tensor2};
auto tensorList = aclCreateTensorList(list1, 2);
aclTensor *list2[] = {tensor4, tensor5};
auto output = aclCreateTensorList(list2, 2);
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
// The AddCustom operator has two inputs (aclTensorList and aclTensor) and one output (aclTensorList).
// Call the first-phase API.
aclnnAddCustomGetWorkspaceSize(tensorList, tensor3, output, &workspaceSize, &executor);
// Set the executor to be reusable.
aclSetAclOpExecutorRepeatable(executor);  
void *addr;
aclSetDynamicTensorAddr(executor, 0, 0, tensorList, addr); // Update the device address of the first aclTensor in the input tensor list.
aclSetDynamicTensorAddr(executor, 0, 1, tensorList, addr); // Update the device address of the second aclTensor in the input tensor list.
aclSetDynamicTensorAddr(executor, 0, 0, output, addr); // Update the device address of the first aclTensor in the output.
aclSetDynamicTensorAddr(executor, 0, 1, output, addr); // Update the device address of the second aclTensor in the output.
...
// Call the second-phase API.
aclnnAddCustom(workspace, workspaceSize, executor, stream);
// Clear the executor.
aclDestroyAclOpExecutor(executor);  

Parent topic: Common APIs