aclnnNLLLossBackward

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: Backward propagation of the negative log-likelihood loss function.
Formula:
- reduction=mean:
  $x\_grad_{target(t)} =\begin{cases} (-gradOutput * w_{(target(t))}) / totalweight &, target(t)=1 \\ 0 &, target(t)=0 \end{cases}$
- reduction=sum:
  $x\_grad_{target(t)} =\begin{cases} -gradOutput * w_{(target(t))} &, target(t)=1 \\ 0 &, target(t)=0 \end{cases}$
- reduction=none:
  $x\_grad_{target(t)} =\begin{cases} -gradOutput_t * w_{(target(t))} &, target(t)=1 \\ 0 &, target(t)=0 \end{cases}$

Function Prototype

Each operator is divided into a two-phase API. First call the "aclnnNLLLossBackwardGetWorkspaceSize" API to obtain the workspace size required for computation and the executor that contains the operator computation process, and then call the "aclnnNLLLossBackward" API to perform the computation.

aclnnStatus aclnnNLLLossBackwardGetWorkspaceSize(
    const aclTensor *gradOutput,
    const aclTensor *self,
    const aclTensor *target,
    const aclTensor *weight,
    int64_t          reduction,
    int64_t          ignoreIndex,
    const aclTensor *totalWeight,
    aclTensor       *out,
    uint64_t        *workspaceSize,
    aclOpExecutor  **executor)

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

aclnnNLLLossBackwardGetWorkspaceSize

Parameters:

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Shape	Non-contiguous Tensor
gradOutput (aclTensor*)	Input	Input aclTensor.	Shape is (N), one-dimensional (with one element), or (). When reduction is 0 and the shape of self is (N, C), the shape supports (N). When reduction is 0 and the shape of self is (C), the shape supports one-dimensional (with one element) or (). When reduction is not 0, the shape supports one-dimensional (with one element) or ().	FLOAT, FLOAT16, BFLOAT16	ND	-	√
self (aclTensor*)	Input	Input aclTensor.	The data type is the same as that of the output parameter out, and the shape is the same as that of out. The shape is (N, C) or (C), where N indicates the batch size and C indicates the number of classes. When the shape of target is (N), the shape of self must be (N, C). When the shape of target is (), the shape of self must be (C).	FLOAT, FLOAT16, BFLOAT16	ND	-	√
target (aclTensor*)	Input	Represents the true labels.	Shape is (N) or (), where each element ranges from 0 to C - 1.	INT64, UINT8, INT32	ND	-	√
weight (aclTensor*)	Input	Represents the scaling weight for each class.	Corresponds to w in the formula, with a shape of (C,).	The data type is consistent with self.	ND	(C)	√
reduction (int64_t)	Input	Specifies the computation mode of the loss function.	Supports 0 (none) \| 1 (mean) \| 2 (sum). 'none' indicates that no reduction is applied. 'mean' indicates that the sum of the output is divided by the number of elements in the output. 'sum' indicates that the output is summed.	INT64	-	-	-
ignoreIndex (int64_t)	Input	Specifies a target value that is ignored and does not affect the input gradient.	-	INT64	-	-	-
totalWeight (aclTensor*)	Input	totalWeight in the formula.	Only when reduction is mean, totalWeight is obtained by taking the weight at the corresponding position through target, then removing the weight corresponding to ignoreIndex, and summing the remaining weights. When reduction is set to other values, this parameter is not processed by default.	The data type is the same as that of weight.	ND	(1,)	-
out (aclTensor*)	Output	Represents out in the formula.	Same shape as self.	Same data type as self.	ND	-	√
workspaceSize (uint64_t*)	Output	Returns the workspace size to be applied for 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 input gradOutput, self, target, weight, out, reduction, or totalWeight is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data type of gradOutput, self, target, weight, or totalWeight is not within the supported range.
		The data types of gradOutput, self, weight, and totalWeight are inconsistent.
		The shape of gradOutput, self, weight, out, or totalWeight is incorrect.
		The reduction value is not within the range of 0 to 2.

aclnnNLLLossBackward

Parameters:

Parameter	Input/Output	Description
workspace	Input	Specifies the workspace memory address applied for on the Device side.
workspaceSize	Input	Specifies the workspace size applied for on the Device side, which is obtained from the first-phase API aclnnNLLLossBackwardGetWorkspaceSize.
executor	Input	Operator executor that contains 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:
- aclnnNLLLossBackward 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_nll_loss_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** tensor) {
  auto size = GetShapeSize(shape) * sizeof(T);
  // Call aclrtMalloc to apply for device-side 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 aclCreateTensor.
  *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.
  // Set 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};
  std::vector<int64_t> selfShape = {2, 3};
  std::vector<int64_t> targetShape = {2};
  std::vector<int64_t> weightShape = {3};
  std::vector<int64_t> totalWeightShape = {1};
  std::vector<int64_t> outShape = {2, 3};
  void* gradOutputDeviceAddr = nullptr;
  void* selfDeviceAddr = nullptr;
  void* targetDeviceAddr = nullptr;
  void* weightDeviceAddr = nullptr;
  void* totalWeightDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* gradOutput = nullptr;
  aclTensor* self = nullptr;
  aclTensor* target = nullptr;
  aclTensor* weight = nullptr;
  aclTensor* totalWeight = nullptr;
  aclTensor* out = nullptr;
  std::vector<float> gradOutputHostData = {3.1, 6.5};
  std::vector<float> selfHostData = {0, 1, 2, 3, 4, 5};
  std::vector<int32_t> targetHostData = {0, 2};
  std::vector<float> weightHostData = {1.1, 1.2, 1.3};
  std::vector<float> totalWeightHostData = {0};
  std::vector<float> outHostData(6, 0);
  int64_t reduction = 0;
  int64_t ignoreIndex = -100;
  // Create the gradOutput aclTensor.
  ret =
      CreateAclTensor(gradOutputHostData, gradOutputShape, &gradOutputDeviceAddr, aclDataType::ACL_FLOAT, &gradOutput);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the self aclTensor.
  ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the target aclTensor.
  ret = CreateAclTensor(targetHostData, targetShape, &targetDeviceAddr, aclDataType::ACL_INT32, &target);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the weight aclTensor.
  ret = CreateAclTensor(weightHostData, weightShape, &weightDeviceAddr, aclDataType::ACL_FLOAT, &weight);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the totalWeight aclTensor.
  ret = CreateAclTensor(totalWeightHostData, totalWeightShape, &totalWeightDeviceAddr, aclDataType::ACL_FLOAT,
                        &totalWeight);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the out aclTensor.
  ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
  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 aclnnNLLLossBackward.
  ret = aclnnNLLLossBackwardGetWorkspaceSize(gradOutput, self, target, weight, reduction, ignoreIndex, totalWeight, out,
                                             &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnNLLLossBackwardGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
  // Allocate device memory based on the workspaceSize computed 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 aclnnNLLLossBackward.
  ret = aclnnNLLLossBackward(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnNLLLossBackward 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 based on the specific API definition.
  auto size = GetShapeSize(outShape);
  std::vector<float> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), outDeviceAddr,
                    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 based on the specific API definition.
  aclDestroyTensor(gradOutput);
  aclDestroyTensor(self);
  aclDestroyTensor(target);
  aclDestroyTensor(weight);
  aclDestroyTensor(totalWeight);
  aclDestroyTensor(out);

  // 7. Release device resources.
  aclrtFree(gradOutputDeviceAddr);
  aclrtFree(selfDeviceAddr);
  aclrtFree(targetDeviceAddr);
  aclrtFree(weightDeviceAddr);
  aclrtFree(totalWeightDeviceAddr);
  aclrtFree(outDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}