aclnnEluBackward

📄 View Source Code

Applicable Products

Function

• Description: Backward computation of the aclnnElu activation function. Outputs the gradient of the forward input of the ELU activation function.

• Formula: $x$ is an element in selfOrResult.

  • When isResult is True:

    $$gradInput = gradOutput * \begin{cases} scale, \quad x > 0\\ inputScale \ast (x + \alpha \ast scale), \quad x \leq 0 \end{cases}$$

  • When isResult is False:

    $$gradInput = gradOutput * \begin{cases} scale, \quad x > 0\\ inputScale \ast \alpha \ast scale \ast exp(x \ast inputScale), \quad x \leq 0 \end{cases}$$

Function Prototype

Each operator is divided into a two-phase API. The "aclnnEluBackwardGetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnEluBackward" API is called to perform the computation.

aclnnStatus aclnnEluBackwardGetWorkspaceSize(
  const aclTensor* gradOutput,
  const aclScalar* alpha,
  const aclScalar* scale,
  const aclScalar* inputScale,
  bool             isResult,
  const aclTensor* selfOrResult,
  aclTensor*       gradInput,
  uint64_t*        workspaceSize,
  aclOpExecutor**  executor)

aclnnStatus aclnnEluBackward(
  void*            workspace,
  uint64_t         workspaceSize,
  aclOpExecutor*   executor,
  aclrtStream      stream)

aclnnEluBackwardGetWorkspaceSize

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  gradOutput (aclTensor*)	Input	Indicates the gradient of the forward output of the ELU activation function, i.e., gradInput in the formula.	Supports empty tensors.	FLOAT, FLOAT16, BFLOAT16	ND	0-8	√
  alpha (aclScalar*)	Input	Indicates the activation coefficient of the ELU activation function, i.e., \alpha in the formula.	If isResult is true, α must be greater than or equal to 0. The data type must be convertible to FLOAT (see Conversion Relationship).	-	-	-	-
  scale (aclScalar*)	Input	Indicates the scaling factor of the ELU activation function, i.e., scale in the formula.	The data type must be convertible to FLOAT (see Conversion Relationship).	-	-	-	-
  inputScale (aclScalar*)	Input.	Indicates the scaling factor of the ELU activation function input, i.e., inputScale in the formula.	The data type must be convertible to FLOAT (see Conversion Relationship).	-	-	-	-
  isResult (bool)	Input	Indicates whether the input passed to the ELU backward computation is the output of the ELU forward computation.	-	-	-	-	-
  selfOrResult (aclTensor*)	Input	When isResult is True, indicates the forward output of the ELU activation function. When isResult is False, indicates the forward input of the ELU activation function.	The data type must be consistent with gradOutput The shape must be consistent with the shape of gradOutput Supports empty tensors.	FLOAT, FLOAT16, BFLOAT16	ND	0-8	√
  gradInput (aclTensor*)	Output	Indicates the gradient of the forward input of the ELU activation function, i.e., the result after deriving the input, which is gradOutput in the formula.	The data type must be a data type convertible from the gradOutput deduction (see Conversion Relationship). The shape must be consistent with the shape of gradOutput	FLOAT, FLOAT16, BFLOAT16	ND	0-8	√
  workspaceSize (uint64_t*)	Output	Returns the workspace size to be applied on the Device side.	-	-	-	-	-
  executor (aclOpExecutor**)	Output	Returns the operator executor, which includes the operator computation flow.	-	-	-	-	-

  • Atlas inference products, Atlas training products: The supported data types are FLOAT and FLOAT16.

• Return value

  aclnnStatus: return code. For details, see aclnn Return Code.

  The first-phase API performs input parameter validation and returns an error in the following scenarios:

  Return Code	Error Code	Description
  ACLNN_ERR_PARAM_NULLPTR	161001	The parameters gradOutput, alpha, scale, inputScale, selfOrResult, and gradInput are null pointers.
  ACLNN_ERR_PARAM_INVALID	161002	The data type of the parameters gradOutput and selfOrResult is not within the supported range.
  		The data types of the parameters gradOutput and selfOrResult are inconsistent.
  		The data type of the parameters alpha, scale, and inputScale cannot be converted to FLOAT.
  		The data type of the parameter gradInput is not convertible from gradOutput
  		The shapes of the parameters gradOutput, selfOrResult, and gradInput are inconsistent.
  		The dimension of the parameters gradOutput, selfOrResult, and gradInput is greater than 8.
  		When the parameter isResult is True, alpha is less than 0.

aclnnEluBackward

• Parameters:

  Parameter	Input/Output	Description
  workspace	Input	Address of the workspace memory applied on the Device side.
  workspaceSize	Input	Size of the workspace applied on the Device side, obtained from the first-phase API aclnnEluBackwardGetWorkspaceSize.
  executor	Input	Operator executor, which includes the operator computation flow.
  stream	Input	Specifies the stream for task execution.

• Return value

  aclnnStatus: return code. For details, see aclnn Return Code.

Constraints

• Deterministic computation:

  • aclnnEluBackward defaults to a deterministic implementation.

Example

The following provides example code for reference only. For details about compilation and running, see Compile and Run Samples.

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_elu_backward.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** selfOrResult) {
  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);

  // Copy host-side data to device-side memory by calling aclrtMemcpy.
  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 the contiguous selfOrResult.
  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];
  }

  // Create an aclTensor by calling the aclCreateTensor API.
  *selfOrResult = 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. See the ACL API.
  // Fill in the deviceId based on the 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.
  std::vector<int64_t> gradOutputShape = {2, 2};
  std::vector<int64_t> selfOrResultShape = {2, 2};
  std::vector<int64_t> gradInputShape = {2, 2};
  void* gradOutputDeviceAddr = nullptr;
  void* selfOrResultDeviceAddr = nullptr;
  void* gradInputDeviceAddr = nullptr;
  aclTensor* gradOutput = nullptr;
  aclScalar* alpha = nullptr;
  aclScalar* scale = nullptr;
  aclScalar* inputScale = nullptr;
  aclTensor* selfOrResult = nullptr;
  aclTensor* gradInput = nullptr;
  std::vector<float> gradOutputHostData = {-2, -1, 0, 1};
  std::vector<float> selfOrResultHostData = {-2, -1, 0, 1};
  std::vector<float> gradInputHostData = {0, 0, 0, 0};
  float alphaValue = 1.0f;
  float scaleValue = 1.0f;
  float inputScaleValue = 1.0f;
  bool isResult = true;
  // Create the gradOutput aclTensor.
  ret = CreateAclTensor(gradOutputHostData, gradOutputShape, &gradOutputDeviceAddr, aclDataType::ACL_FLOAT,
                        &gradOutput);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the alpha aclScalar.
  alpha = aclCreateScalar(&alphaValue, aclDataType::ACL_FLOAT);
  CHECK_RET(alpha != nullptr, return ret);
  // Create the scale aclScalar.
  scale = aclCreateScalar(&scaleValue, aclDataType::ACL_FLOAT);
  CHECK_RET(scale != nullptr, return ret);
  // Create the inputScale aclScalar.
  inputScale = aclCreateScalar(&inputScaleValue, aclDataType::ACL_FLOAT);
  CHECK_RET(inputScale != nullptr, return ret);
  // Create selfOrResult aclTensor.
  ret = CreateAclTensor(selfOrResultHostData, selfOrResultShape, &selfOrResultDeviceAddr, aclDataType::ACL_FLOAT,
                        &selfOrResult);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create gradInput aclTensor.
  ret = CreateAclTensor(gradInputHostData, gradInputShape, &gradInputDeviceAddr, aclDataType::ACL_FLOAT, &gradInput);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;

  // Example of aclnnEluBackward API call.
  // 3. Call the CANN operator library API. Replace it with the specific API name.
  // Call the first-phase API of aclnnEluBackward.
  ret = aclnnEluBackwardGetWorkspaceSize(gradOutput, alpha, scale, inputScale, isResult, selfOrResult, gradInput,
                                         &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnEluBackwardGetWorkspaceSize 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 aclnnEluBackward.
  ret = aclnnEluBackward(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnEluBackward 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 side memory to the host side. Modify this based on the specific API definition.
  auto size = GetShapeSize(gradInputShape);
  std::vector<float> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), gradInputDeviceAddr,
                    size * sizeof(resultData[0]), 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 < size; i++) {
    LOG_PRINT("result[%ld] is: %f\n", i, resultData[i]);
  }

  // 6. Release aclTensor and aclScalar. Modify this based on the specific API definition.
  aclDestroyTensor(gradOutput);
  aclDestroyScalar(alpha);
  aclDestroyScalar(scale);
  aclDestroyScalar(inputScale);
  aclDestroyTensor(selfOrResult);
  aclDestroyTensor(gradInput);

  // 7. Release device resources. Modify based on the specific API definition.
  aclrtFree(gradOutputDeviceAddr);
  aclrtFree(selfOrResultDeviceAddr);
  aclrtFree(gradInputDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}