aclnnSmoothL1Loss

Applicable Products

Product	Supported
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: Computes the SmoothL1 loss function.
Formula:

For a batch of size N, when reduction is none, the loss function is defined as:
$\ell(self,target) = L = \{l_1,\dots,l_N\}^\top$
where $l_n$ is:
$l_n = \begin{cases} 0.5(self_n-target_n)^2/beta, & if |self_n-target_n| < beta \\ |self_n-target_n| - 0.5*beta, & otherwise \end{cases}$
When reduction is mean or sum,
$\ell(self,target)=\begin{cases} mean(L), & \text{if reduction} = \text{mean}\\ sum(L), & \text{if reduction} = \text{sum} \end{cases}$

Function Prototype

Each operator is divided into a two-phase API. The "aclnnSmoothL1LossGetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnSmoothL1Loss" API is called to perform the computation.

aclnnStatus aclnnSmoothL1LossGetWorkspaceSize(
      const aclTensor* self,
      const aclTensor* target,
      int64_t reduction,
      float beta,
      aclTensor* result,
      uint64_t* workspaceSize,
      aclOpExecutor** executor)

aclnnStatus aclnnSmoothL1Loss(
    void* workspace,
    uint64_t workspaceSize,
    aclOpExecutor* executor,
    aclrtStream stream)

aclnnSmoothL1LossGetWorkspaceSize

Parameters:

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
self (aclTensor*)	Input	Input tensor, which is the input self in the formula.	The shape must satisfy the broadcast relationship with target. The data type must comply with the deduction rules with the target data type (see Conversion Relationship).	FLOAT, FLOAT16, BFLOAT16	ND, NCL, NCHW, NHWC	1-8	√
target (aclTensor*)	Input	Ground-truth label, i.e., the input y in the formula.	The shape must satisfy the broadcast relationship with self. The data type must comply with the deduction rules with self (see Conversion Relationship).	FLOAT, FLOAT16, BFLOAT16	ND, NCL, NCHW, NHWC	1-8	√
reduction (int64_t)	Input	Specifies the reduction to apply to the output, i.e., the parameter reduction in the formula.	Supports 0 (none) \| 1 (mean) \| 2 (sum). 'none' indicates no reduction is applied. 'mean' indicates the sum of the output will be divided by the number of elements in the output. 'sum' indicates the output will be summed.	INT64	-	-	-
beta (float)	Input	Specifies the threshold value for switching between L1 and L2 loss.	The value must be non-negative.	FLOAT	-	-	-
result (aclTensor*)	Output	The loss function ℓ(self, target) output in the formula.	When reduction is none, the shape matches the broadcast result of self and target. When reduction is mean or sum, the shape is [1]. The data type is the convertible data type derived from self and target (see conversion relationship).	FLOAT, FLOAT16, BFLOAT16	ND, NCL, NCHW, NHWC	1-8	√
workspaceSize (uint64_t*)	Output	Returns the workspace size to be allocated on the Device side.	-	-	-	-	-
executor (aclOpExecutor**)	Output	Returns the op executor, which contains the operator computation flow.	-	-	-	-	-

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 Value	Error Code	Description
ACLNN_ERR_PARAM_NULLPTR	161001	The passed self, target, or result is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data type of self, target, or result is not within the supported range.
		The shape of self, target, or result does not meet the constraints.
		The reduction parameter does not meet the constraints.
		The beta parameter does not meet the constraints.
		The data types of self and target do not satisfy the data type deduction rules.
		The shapes of self and target do not satisfy the broadcast relationship.
		When reduction is 0, the shape of result is inconsistent with the broadcast shape of self and target.

aclnnSmoothL1Loss

Parameters:

Parameter	Input/Output	Description
workspace	Input	Workspace memory address applied for on the Device side.
workspaceSize	Input	Workspace size applied for on the Device side, obtained through the first-phase API aclnnSmoothL1LossGetWorkspaceSize.
executor	Input	Op executor that contains the operator computation process.
stream	Input	Specifies the stream for task execution.

Return value

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

Constraints

Deterministic computation:
aclnnSmoothL1Loss 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_smooth_l1_loss.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 device memory.
  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 data from the host to the device 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 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];
  }

  // Create an aclTensor by calling the aclCreateTensor API.
  *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. See the acl API reference.
  // Set 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.
  std::vector<int64_t> selfShape = {2, 2, 7, 7};
  std::vector<int64_t> targetShape = {2, 2, 7, 7};
  std::vector<int64_t> resultShape = {2, 2, 7, 7};

  // Create the self aclTensor.
  std::vector<float> selfData(GetShapeSize(selfShape)* 2, 1);
  aclTensor* self = nullptr;
  void *selfDeviceAddr = nullptr;
  ret = CreateAclTensor(selfData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT16, &self);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the target aclTensor.
  std::vector<float> targetData(GetShapeSize(targetShape)* 2, 1);
  aclTensor* target = nullptr;
  void *targetDeviceAddr = nullptr;
  ret = CreateAclTensor(targetData, targetShape, &targetDeviceAddr, aclDataType::ACL_FLOAT16, &target);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the result aclTensor.
  std::vector<float> resultData(GetShapeSize(resultShape)* 2, 1);
  aclTensor* result = nullptr;
  void *resultDeviceAddr = nullptr;
  ret = CreateAclTensor(resultData, resultShape, &resultDeviceAddr, aclDataType::ACL_FLOAT16, &result);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  // 3. Call the CANN operator library API. Replace with the specific API name.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // Call the first-phase API of aclnnSmoothL1Loss.
  int64_t reduction = 0;
  float beta = 1.0;
  ret = aclnnSmoothL1LossGetWorkspaceSize(self, target, reduction, beta, result, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnSmoothL1LossGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
  // Allocate 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 aclnnSmoothL1Loss.
  ret = aclnnSmoothL1Loss(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnSmoothL1Loss 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, copy the result from the device memory to the host side, and modify as needed based on the specific API definition.
  auto size = GetShapeSize(resultShape);
  std::vector<float> resultOutData(size, 0);
  ret = aclrtMemcpy(resultOutData.data(), resultOutData.size() * sizeof(resultOutData[0]), resultDeviceAddr,
                    size * sizeof(resultOutData[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, resultOutData[i]);
  }

  // 6. Release aclTensor and aclScalar. Modify based on the specific API definition.
  aclDestroyTensor(self);
  aclDestroyTensor(target);
  aclDestroyTensor(result);
  // 7. Release device resources. Set parameters based on the specific API definition.
  aclrtFree(selfDeviceAddr);
  aclrtFree(targetDeviceAddr);
  aclrtFree(resultDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();
  return 0;
}