Documentation
Download
文档中心
aclnnObfuscationCalculateV2

aclnnObfuscationCalculateV2

Applicable Products

ProductSupported
Ascend 950PR/Ascend 950DT×
Atlas A3 training products/Atlas A3 inference products×
Atlas A2 training products/Atlas A2 inference products√
Atlas 200I/500 A2 inference products×
Atlas inference products√
Atlas training products×

Function

  • Description: Sends the tensor x and configuration parameters (such as param and cmd) to the PMCC obfuscation engine. The CA module of the engine invokes the TA module to perform tensor obfuscation processing, and finally returns an obfuscated tensor y with the same shape as x.

  • Background: The PMCC (Privacy & Model Confidential Computing) model obfuscation feature uses the TrustZone trusted execution environment in the CPU core to isolate and store obfuscation factors, derive obfuscation masks, and perform dynamic mask addition. PMCC builds a model obfuscation engine CA (Client Application in the normal OS) and a model obfuscation engine TA (Trusted Application in the TEE OS) based on NPU TrustZone. 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. This API adds obfCoefficient, which is introduced in the full DeepSeek version. To ensure that the model obfuscation performance meets requirements, an obfuscation coefficient is added to process the input data proportionally according to the obfuscation coefficient.

Function Prototype

Each operator is divided into a two-phase API. You must first call the "aclnnObfuscationCalculateV2GetWorkspaceSize" API to obtain the workspace size required for computation and the executor that contains the operator computation process, and then call the "aclnnObfuscationCalculateV2" API to execute the computation.

aclnnStatus aclnnObfuscationCalculateV2GetWorkspaceSize(
  int32_t          fd,
  const aclTensor *x,
  int32_t          param,
  int32_t          cmd,
  float            obfCoefficient,
  aclTensor       *y,
  uint64_t        *workspaceSize,
  aclOpExecutor  **executor)
aclnnStatus aclnnObfuscationCalculateV2(
  void          *workspace,
  uint64_t       workspaceSize,
  aclOpExecutor *executor,
  aclrtStream    stream)

aclnnObfuscationCalculateV2GetWorkspaceSize

  • Parameters

    Parameter Input/Output Description Instruction Data Type Data Format Dimension (Shape) Non-contiguous Tensor
    fd (int32_t) Input Socket connector. Specifies fd[0] from the output of aclnnObfuscationSetupV2 during resource initialization. INT32 ND - -
    x (const aclTensor*) Input Tensor to be obfuscated. Empty tensors are not supported. FLOAT, FLOAT16, INT8, BFLOAT16 ND [...,H] ×
    param (int32_t) Input Reserved parameter field. Only 0 is supported in the current version. INT32 - - -
    cmd (int32_t) Input Instruction number of the obfuscation operator. Only 1 is supported in the current version. INT32 - - -
    obfCoefficient (float) Input Obfuscation coefficient used for obfuscation processing. Value range: (0.0, 1.0]. - - - -
    y (aclTensor*) Output Tensor after obfuscation processing. Data type and shape are the same as those of x. FLOAT, FLOAT16, INT8, BFLOAT16 ND [...,H] ×
    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. - - - - -

  • Atlas inference products: BFLOAT16 is not supported.

  • 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 x or y is a null pointer.
    ACLNN_ERR_PARAM_INVALID 161002 The data type or data format of x is not within the supported range.
    The data types of x and y are inconsistent.
    The shapes of x and y are inconsistent.

aclnnObfuscationCalculateV2

  • Parameters

    Parameter Input/Output Description
    workspace Input Address of the workspace memory applied for on the Device side.
    workspaceSize Input Size of the workspace applied for on the Device side, obtained through the first-phase API aclnnObfuscationCalculateV2GetWorkspaceSize.
    executor Input Operator executor that contains the operator computation flow.
    stream Input Specifies the stream that executes the task.

  • Return Value

Returns the aclnnStatus status code. For details, see aclnn Return Codes.

Constraints

  • Deterministic computation:

    • aclnnObfuscationCalculateV2 defaults to a deterministic implementation.

  • This API is used together with aclnnObfuscationSetupV2 to implement the PMCC model obfuscation function. The usage is as follows:

    • First, call aclnnObfuscationSetupV2 for resource initialization. This API can be called repeatedly, and the last initialization takes effect.

    • Then, call aclnnObfuscationCalculateV2 multiple times for tensor obfuscation processing.

    • Finally, call aclnnObfuscationSetupV2 for resource release. This API can be called only once. Alternatively, you can release resources by terminating the program process instead of explicitly calling the API.

Example

The sample code is as follows, 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_v2.h"
#include "aclnnop/aclnn_obfuscation_calculate_v2.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, 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;
  float obfCoefficient = 0.5;
  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 aclnnObfuscationSetupV2.
  ret = aclnnObfuscationSetupV2GetWorkspaceSize(fdToClose, dataType, hiddenSize, tpRank, modelObfSeedId,dataObfSeedId, cmd, threadNum, obfCoefficient, fd, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationSetupV2GetWorkspaceSize 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 aclnnObfuscationSetupV2 second-phase API.
  ret = aclnnObfuscationSetupV2(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationSetupV2 failed. ERROR: %d\n", ret); return ret);

  // 4. (Boilerplate) Synchronously wait for the 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 step 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 it with the specific API.
  uint64_t workspaceSize2 = 0;
  aclOpExecutor *executor2;

  // Call the first-phase API of aclnnObfuscationCalculateV2.
  ret = aclnnObfuscationCalculateV2GetWorkspaceSize(fdInput, x, param, cmd2, obfCoefficient, y, &workspaceSize2, &executor2);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationCalculateV2GetWorkspaceSize 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 aclnnObfuscationCalculateV2.
  ret = aclnnObfuscationCalculateV2(workspaceAddr2, workspaceSize2, executor2, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationCalculateV2 failed. ERROR : %d\n", ret); return ret);
  // 8. 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);

  // 9. Obtain the output value. y represents the data after obfuscation processing. Copy the result from the device-side memory to the host side. Modify 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 the input and output. 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 with the specific API name.
  uint64_t workspaceSize3 = 0;
  aclOpExecutor* executor3;

  // Call the first-phase API of aclnnObfuscationSetupV2.
  ret = aclnnObfuscationSetupV2GetWorkspaceSize(fdToClose, dataType, hiddenSize, tpRank, modelObfSeedId,dataObfSeedId, cmd, threadNum, obfCoefficient, fd, &workspaceSize3, &executor3);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationSetupV2GetWorkspaceSize failed. ERROR: %d\n", ret); return ret);

  // Apply for 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 aclnnObfuscationSetupV2.
  ret = aclnnObfuscationSetupV2(workspaceAddr3, workspaceSize3, executor3, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnObfuscationSetupV2 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 ObfuscationCalculateV2 API. Modify based on the specific API definition.
  aclDestroyTensor(x);
  aclDestroyTensor(y);
  aclDestroyTensor(fd);

  // 14. Release the device resources involved in the ObfuscationCalculateV2 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 ObfuscationSetupV2 API.
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}
This site uses cookies to improve your browsing experience. By continuing to browse this site, you agree to our use of cookies.View Details