aclnnObfuscationSetup
Applicable Products
| Product | Supported |
|---|---|
| × | |
| × | |
| √ | |
| × | |
| √ | |
| × |
Function
Description: Completes resource initialization and release for the PMCC model obfuscation engine.
Resource initialization: Establishes a socket connection with the PMCC obfuscation engine CA, initializes the CA and TA, and returns a socket connector.
Resource release: Disconnects the socket connection from the PMCC obfuscation engine CA.
Background: The PMCC (Privacy & Model Confidential Computing) model obfuscation feature uses the TrustZone trusted execution environment in CPU cores to isolate and store obfuscation factors, derive obfuscation masks, and perform dynamic mask addition. Based on NPU TrustZone, PMCC builds the model obfuscation engine CA (Client Application in the normal OS) and the model obfuscation engine TA (Trusted Application in the TEE OS). To enable the model to access the model obfuscation engine TA during inference execution, the AICPU operator mechanism and the localhost socket within the NPU card are used for relay.
Function Prototype
Each operator is divided into a two-phase API. The "aclnnObfuscationSetupGetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnObfuscationSetup" API is called to perform the computation.
aclnnStatus aclnnObfuscationSetupGetWorkspaceSize(
int32_t fdToClose,
int32_t dataType,
int32_t hiddenSize,
int32_t tpRank,
int32_t modelObfSeedId,
int32_t dataObfSeedId,
int32_t cmd,
int32_t threadNum,
aclTensor *fd,
uint64_t *workspaceSize,
aclOpExecutor **executor)aclnnStatus aclnnObfuscationSetup(
void *workspace,
uint64_t workspaceSize,
aclOpExecutor *executor,
aclrtStream stream)aclnnObfuscationSetupGetWorkspaceSize
Parameters
Parameter Input/Output Description Instruction Data Type Data Format Dimension (Shape) Non-Contiguous Tensor fdToClose (int32_t) Input Socket connector to be closed. When cmd is 3, fill in the fd returned by this operator when cmd is 1; otherwise, fill in 0. INT32 - - - dataType (int32_t) Input Represents the number of the tensor data type. - Only when cmd is set to 1 or 2, a valid value is required; otherwise, set it to 0.
Atlas inference products : Select from {0, 1}, where 0 indicates FLOAT and 1 indicates FLOAT16.Atlas A2 training products/Atlas A2 inference products : Select from {0, 1, 2, 27}, where 0 indicates FLOAT, 1 indicates FLOAT16, 2 indicates INT8, and 27 indicates BF16.
INT32 - - - hiddenSize (int32_t) Input Hidden layer dimension. Specifies a valid value only when cmd is set to 1 or 2; otherwise, set it to 0. The value ranges from 1 to 10000. INT32 - - - tpRank (int32_t) Input TP Rank. Supports values from 0 to 1024. A valid value is required only when cmd is set to 1 or 2; otherwise, set it to 0. INT32 - - - modelObfSeedId (int32_t) Input ID of the model obfuscation factor. Used by the TA to query the model obfuscation factor from the TEE KMC. A valid value is required only when cmd is set to 1 or 2; otherwise, set it to 0. INT32 - - - dataObfSeedId (int32_t) Input Data obfuscation factor ID. Used by the TA to query the data obfuscation factor from the TEE KMC. Specify a valid value only when cmd is set to 1 or 2; otherwise, set it to 0. INT32 - - - cmd (int32_t) Input Setup instruction number. Selects from {1, 2, 16}. When set to 1, performs resource initialization in normal mode; when set to 2, performs resource initialization in high-precision mode; when set to 16, performs resource release. INT32 - - - threadNum (int32_t) Input Number of threads used by CA/TA for obfuscation processing. Select from {1, 2, 3, 4, 5, 6}. A valid value is required only when cmd is set to 1 or 2; otherwise, set to 0. INT32 - - - fd (aclTensor*) Output Socket connector. Empty tensors are not supported. INT32 ND 1 × workspaceSize (uint64_t*) Output Returns the workspace size that the user needs to apply for on the Device side. - - - - - executor (aclOpExecutor**) Output Returns the op executor, which contains the operator computation flow. - - - - - Return Value
aclnnStatus: return status code. For details, see aclnn Return Codes.
The first-phase API performs input parameter validation and returns an error in the following scenarios:
Return Value Error Code Description ACLNN_ERR_PARAM_NULLPTR 161001 The input fd is a null pointer. ACLNN_ERR_PARAM_INVALID 161002 The data type or data format of fd is not within the supported range.
aclnnObfuscationSetup
Parameters
Parameter Input/Output Description workspace Input Specifies the workspace memory address applied on the Device side. workspaceSize Input Specifies the workspace size applied on the Device side, obtained through the first-phase API aclnnObfuscationSetupGetWorkspaceSize. executor Input Operator executor that contains the operator computation flow. stream Input Specifies the stream for task execution. Return Value
Returns an aclnnStatus return code. For details, see aclnn Return Codes.
Constraints
Deterministic computation:
- aclnnObfuscationSetup defaults to a deterministic implementation.
This API is used together with aclnnObfuscationCalculate to implement the PMCC model obfuscation function. The usage is as follows:
First, call aclnnObfuscationSetup for resource initialization. This API can be called repeatedly, and the last initialization takes effect.
Then, call aclnnObfuscationCalculate multiple times for tensor obfuscation.
Finally, call aclnnObfuscationSetup for resource release. This can be called only once. Alternatively, you can release resources by terminating the program process instead of explicitly calling the API.
Example
The following is a sample code for reference only. For details about compilation and execution, see Compile and Run Samples.
#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_obfuscation_setup.h"
#include "aclnnop/aclnn_obfuscation_calculate.h"
#define CHECK_RET(cond, return_expr) \
do { \
if (!(cond)) { \
return_expr; \
} \
} while (0)
#define LOG_PRINT(message, ...) \
do { \
printf(message, ##__VA_ARGS__); \
} while (0)
int64_t GetShapeSize(const std::vector<int64_t>& shape) {
int64_t shapeSize = 1;
for (auto i : shape) {
shapeSize *= i;
}
return shapeSize;
}
int Init(int32_t deviceId, aclrtStream* stream) {
// Boilerplate for resource initialization.
auto ret = aclInit(nullptr);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret); return ret);
ret = aclrtSetDevice(deviceId);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret);
ret = aclrtCreateStream(stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret); return ret);
return 0;
}
template <typename T>
int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
aclDataType dataType, aclTensor** tensor) {
auto size = GetShapeSize(shape) * sizeof(T);
// Call aclrtMalloc to apply for memory on the device side.
auto ret = aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMalloc failed. ERROR: %d\n", ret); return ret);
// Call aclrtMemcpy to copy data from the host side to the device memory.
ret = aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMemcpy failed. ERROR: %d\n", ret); return ret);
// Calculate the strides of a contiguous tensor.
std::vector<int64_t> strides(shape.size(), 1);
for (int64_t i = shape.size() - 2; i >= 0; i--) {
strides[i] = shape[i + 1] * strides[i + 1];
}
// Call the aclCreateTensor API to create an aclTensor.
*tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
shape.data(), shape.size(), *deviceAddr);
return 0;
}
int main() {
// 1. (Boilerplate) Initialize the device/stream. Refer to the acl API manual.
// Fill in the deviceId based on your actual device.
int32_t deviceId = 0;
aclrtStream stream;
auto ret = Init(deviceId, &stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
// 2. Construct the input and output. Customize the construction based on the API interface.
std::vector<int64_t> fdShape = {1};
void* fdDeviceAddr = nullptr;
int32_t fdToClose = 0;
int32_t dataType = 0;
int32_t hiddenSize = 4;
int32_t tpRank = 0;
int32_t modelObfSeedId = 123456;
int32_t dataObfSeedId = 654321;
int32_t cmd = 1;
int32_t threadNum = 4;
aclTensor* fd = nullptr;
std::vector<float> fdHostData = {-1};
//Create the fd aclTensor.
ret = CreateAclTensor(fdHostData, fdShape, &fdDeviceAddr, aclDataType::ACL_INT32, &fd);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 3. Call the CANN operator library API. Replace it with the specific API name.
uint64_t workspaceSize = 0;
aclOpExecutor* executor;
// Call the first-phase API of aclnnObfuscationSetup.
ret = aclnnObfuscationSetupGetWorkspaceSize(fdToClose, dataType, hiddenSize, tpRank, modelObfSeedId,dataObfSeedId, cmd, threadNum, fd, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationSetupGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
// Apply for device memory based on the workspaceSize calculated by the first-phase API.
void* workspaceAddr = nullptr;
if (workspaceSize > 0) {
ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
}
// Call the second-phase API of aclnnObfuscationSetup.
ret = aclnnObfuscationSetup(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationSetup failed. ERROR: %d\n", ret); return ret);
// 4. (Boilerplate) Synchronize and wait for task execution to complete.
ret = aclrtSynchronizeStream(stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
// 5. Obtain the output value and copy the result from the device memory to the host memory. Modify this based on the specific API definition.
auto fdSize = GetShapeSize(fdShape);
std::vector<int32_t> fdData(fdSize, 0);
ret = aclrtMemcpy(fdData.data(), fdData.size() * sizeof(fdData[0]), fdDeviceAddr, fdSize * sizeof(int32_t),ACL_MEMCPY_DEVICE_TO_HOST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR : %d\n", ret); return ret);
for (int64_t i = 0; i < fdSize; i++) {
LOG_PRINT("fdData[%ld] is : %d\n", i, fdData[i]);
}
// 6. Construct the input and output. Construct them based on the API definition.
std::vector<int64_t> xShape = {2, 4};
std::vector<int64_t> yShape = {2, 4};
void *xDeviceAddr = nullptr;
void *yDeviceAddr = nullptr;
int32_t fdInput = fdData[0];
int32_t param = 4;
aclTensor *x = nullptr;
int32_t cmd2 = 1;
aclTensor *y = nullptr;
std::vector<float> xHostData = {0.86, 0.79, 0.43, 0.37, 0.51, 0.89, 0.34, 0.49};
std::vector<float> yHostData = {0, 0, 0, 0, 0, 0, 0, 0};
// Create x aclTensor.
ret = CreateAclTensor(xHostData, xShape, &xDeviceAddr, aclDataType::ACL_FLOAT, &x);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create y aclTensor.
ret = CreateAclTensor(yHostData, yShape, &yDeviceAddr, aclDataType::ACL_FLOAT, &y);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 7. Call the CANN operator library API. Replace with the specific API.
uint64_t workspaceSize2 = 0;
aclOpExecutor *executor2;
// Call the first-phase API of aclnnObfuscationCalculate.
ret = aclnnObfuscationCalculateGetWorkspaceSize(fdInput, x, param, cmd2, y, &workspaceSize2, &executor2);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationCalculateGetWorkspaceSize failed. ERROR : %d\n",ret); return ret);
// Apply for device memory based on the workspaceSize calculated by the first-phase API.
void *workspaceAddr2 = nullptr;
if (workspaceSize2 > 0) {
ret = aclrtMalloc(&workspaceAddr2, workspaceSize2, ACL_MEM_MALLOC_HUGE_FIRST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
}
// Call the second-phase API of aclnnObfuscationCalculate.
ret = aclnnObfuscationCalculate(workspaceAddr2, workspaceSize2, executor2, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationCalculate failed. ERROR : %d\n", ret); return ret);
// 8. Boilerplate: synchronize and wait for task execution to complete.
ret = aclrtSynchronizeStream(stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR : %d\n", ret); return ret);
// 9. Obtain the output value. y indicates the obfuscated data. Copy the result from the device-side memory to the host side. Modify this based on the specific API definition.
// y
auto ySize = GetShapeSize(yShape);
std::vector<float> yData(ySize, 0);
ret = aclrtMemcpy(yData.data(), yData.size() * sizeof(yData[0]), yDeviceAddr, ySize * sizeof(float),ACL_MEMCPY_DEVICE_TO_HOST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR : %d\n", ret); return ret);
for (int64_t i = 0; i < ySize; i++) {
LOG_PRINT("yData[%ld] is : %f\n", i, yData[i]);
}
// 10. Construct inputs and outputs. Customize the construction based on the API.
fdToClose = fdInput;
dataType = 0;
hiddenSize = 0;
tpRank = 0;
modelObfSeedId = 0;
dataObfSeedId = 0;
cmd = 16;
threadNum = 4;
// 11. Call the CANN operator library API. Replace this with the specific API name.
uint64_t workspaceSize3 = 0;
aclOpExecutor* executor3;
// Call the first-phase API of aclnnObfuscationSetup.
ret = aclnnObfuscationSetupGetWorkspaceSize(fdToClose, dataType, hiddenSize, tpRank, modelObfSeedId,dataObfSeedId, cmd, threadNum, fd, &workspaceSize3, &executor3);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationSetupGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
// Allocate device memory based on the workspaceSize calculated by the first-phase API.
void* workspaceAddr3 = nullptr;
if (workspaceSize3 > 0) {
ret = aclrtMalloc(&workspaceAddr3, workspaceSize3, ACL_MEM_MALLOC_HUGE_FIRST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
}
// Call the second-phase API of aclnnObfuscationSetup.
ret = aclnnObfuscationSetup(workspaceAddr3, workspaceSize3, executor3, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationSetup failed. ERROR: %d\n", ret); return ret);
// 12. (Boilerplate) Synchronously wait for task execution to complete.
ret = aclrtSynchronizeStream(stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
// 13. Release the aclTensor and aclScalar involved in the ObfuscationCalculate API. Modify based on the specific API definition.
aclDestroyTensor(x);
aclDestroyTensor(y);
aclDestroyTensor(fd);
// 14. Release the device resources involved in the ObfuscationCalculate API. Modify based on the specific API definition.
aclrtFree(xDeviceAddr);
aclrtFree(yDeviceAddr);
aclrtFree(fdDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
if (workspaceSize2 > 0) {
aclrtFree(workspaceAddr2);
}
if (workspaceSize3 > 0) {
aclrtFree(workspaceAddr3);
}
// 15. Release the device resources involved in the ObfuscationSetup API.
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}