aclnnSyncBatchNormGatherStats

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:

Collects the mean and variance from all devices and updates the global mean and variance.

Formula: $batchMean = \frac{\sum^N_{i=0}{totalSum[i]}}{\sum^N_{i=0}{sampleCount[i]}}$ $batchVar = \frac{\sum^N_{i=0}{totalSquareSum[i]}}{\sum^N_{i=0}{sampleCount[i]}} - batchMean^2$ $batchInvstd = \frac{1}{\sqrt{batchVar + ε}}$ $runningMean = runningMean*(1-momentum) + momentum*batchMean$ $runningVar = runningVar*(1-momentum) + momentum*(batchVar* \frac{\sum^N_{i=0} {sampleCount[i]}}{\sum^N_{i=0}{sampleCount[i]}-1})$

Function Prototype

Each operator is divided into a two-phase API. You must first call the "aclnnSyncBatchNormGatherStatsGetWorkspaceSize" API to obtain the input parameters and calculate the required workspace size based on the computation process, and then call the "aclnnSyncBatchNormGatherStats" API to perform the computation.

aclnnStatus aclnnSyncBatchNormGatherStatsGetWorkspaceSize(
  const aclTensor   *totalSum,
  const aclTensor   *totalSquareSum,
  const aclTensor   *sampleCount,
  aclTensor         *mean,
  aclTensor         *variance,
  float              momentum,
  float              eps,
  aclTensor         *batchMean,
  aclTensor         *batchInvstd,
  uint64_t          *workspaceSize,
  aclOpExecutor     **executor)

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

aclnnSyncBatchNormGatherStatsGetWorkspaceSize

Parameters

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (shape)	Non-contiguous Tensor
totalSum (aclTensor*)	Input	Indicates the sum of channel features of each device, corresponding to totalSum in the formula.	The first dimension must be greater than 0.	BFLOAT16, FLOAT16, FLOAT	ND	2	√
totalSquareSum (aclTensor*)	Input	Indicates the channel feature squares of each device, corresponding to totalSquareSum in the formula.	The first dimension must be greater than 0. The shape is the same as that of totalSum.	BFLOAT16, FLOAT16, FLOAT	ND	2	√
sampleCount (aclTensor*)	Input	Indicates the sample count of each device, corresponding to sampleCount in the formula.	The first dimension must be greater than 0. The shape must be consistent with the first dimension of totalSum.	BFLOAT16, FLOAT16, FLOAT, INT32	ND	1	√
mean (aclTensor*)	Input	Indicates the mean value during computation, corresponding to runningMean in the formula.	Supports empty tensors. The shape is consistent with the second dimension of totalSum.	BFLOAT16, FLOAT16, FLOAT	ND	1	√
variance (aclTensor*)	Input	Indicates the variance during computation, corresponding to runningVar in the formula.	Supports empty tensors. The shape is consistent with the second dimension of totalSum.	BFLOAT16, FLOAT16, FLOAT	ND	1	√
momentum (float)	Input	Exponential smoothing parameter for runningMean and runningVar.	-	-	-	-	-
eps (float)	Input	An offset used to prevent division by zero.	-	-	-	-	-
batchMean (aclTensor*)	Output	Represents the global batch mean, corresponding to batchMean in the formula.	The first dimension must be greater than 0.	BFLOAT16, FLOAT16, FLOAT	ND	1	√
batchInvstd (aclTensor*)	Output	Indicates the reciprocal of the standard deviation, corresponding to batchInvstd in the formula.	The first dimension must be greater than 0.	BFLOAT16, FLOAT16, FLOAT	ND	1	√
workspaceSize (uint64_t*)	Output	Returns the workspace size to be applied on the Device side.	-	-	-	-	-
executor (aclOpExecutor**)	Output	Returns the operator executor, which contains the operator computation flow.	-	-	-	-	-

Atlas A3 training products/Atlas A3 inference products, Atlas A2 training products/Atlas A2 inference products: The data type of the parameters totalSum, totalSquareSum, mean, variance, batchMean, and batchInvstd does not support BFLOAT16.

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 input totalSum, totalSquareSum, sampleCount, mean, variance, batchMean, or batchInvstd is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data type of input totalSum, totalSquareSum, sampleCount, mean, variance, batchMean, or batchInvstd is not within the supported range.
		The data format of input totalSum, totalSquareSum, sampleCount, mean, or variance is not within the supported range.
		The shape of input totalSum, totalSquareSum, sampleCount, mean, variance, batchMean, or batchInvstd is not within the supported range.

aclnnSyncBatchNormGatherStats

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 through the first-phase API aclnnSyncBatchNormGatherStatsGetWorkspaceSize.
executor	Input	Operator executor, which 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:
- aclnnSyncBatchNormGatherStats 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_sync_batch_norm_gather_stats.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 side 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);

  // Compute 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. For details, see 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> totalSumShape = {1, 2};
  std::vector<int64_t> totalSquareSumShape = {1, 2};
  std::vector<int64_t> sampleCountShape = {1};
  std::vector<int64_t> meanShape = {2};
  std::vector<int64_t> varShape = {2};
  std::vector<int64_t> batchMeanShape = {2};
  std::vector<int64_t> batchInvstdShape = {2};
  void* totalSumDeviceAddr = nullptr;
  void* totalSquareSumDeviceAddr = nullptr;
  void* sampleCountDeviceAddr = nullptr;
  void* meanDeviceAddr = nullptr;
  void* varDeviceAddr = nullptr;
  void* batchMeanDeviceAddr = nullptr;
  void* batchInvstdDeviceAddr = nullptr;
  aclTensor* totalSum = nullptr;
  aclTensor* totalSquareSum = nullptr;
  aclTensor* sampleCount = nullptr;
  aclTensor* mean = nullptr;
  aclTensor* var = nullptr;
  aclTensor* batchMean = nullptr;
  aclTensor* batchInvstd = nullptr;
  std::vector<float> totalSumData = {300, 400};
  std::vector<float> totalSquareSumData = {300, 400};
  std::vector<int32_t> sampleCountData = {400};
  std::vector<float> meanData = {400, 400};
  std::vector<float> varData = {400, 400};
  std::vector<float> batchMeanData = {0, 0};
  std::vector<float> batchInvstdData = {0, 0};
  float momentum = 1e-1;
  float eps = 1e-5;
  // Create the input totalSum aclTensor.
  ret = CreateAclTensor(totalSumData, totalSumShape, &totalSumDeviceAddr, aclDataType::ACL_FLOAT, &totalSum);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the input totalSquareSum aclTensor.
  ret = CreateAclTensor(totalSquareSumData, totalSquareSumShape, &totalSquareSumDeviceAddr, aclDataType::ACL_FLOAT, &totalSquareSum);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the input sampleCount aclTensor.
  ret = CreateAclTensor(sampleCountData, sampleCountShape, &sampleCountDeviceAddr, aclDataType::ACL_INT32, &sampleCount);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the input meanData aclTensor.
  ret = CreateAclTensor(meanData, meanShape, &meanDeviceAddr, aclDataType::ACL_FLOAT, &mean);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the input varData aclTensor.
  ret = CreateAclTensor(varData, varShape, &varDeviceAddr, aclDataType::ACL_FLOAT, &var);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the input batchMeanData aclTensor.
  ret = CreateAclTensor(batchMeanData, batchMeanShape, &batchMeanDeviceAddr, aclDataType::ACL_FLOAT, &batchMean);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the input batchInvstdData aclTensor.
  ret = CreateAclTensor(batchInvstdData, batchInvstdShape, &batchInvstdDeviceAddr, aclDataType::ACL_FLOAT, &batchInvstd);
  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 aclnnSyncBatchNormGatherStats.
  ret = aclnnSyncBatchNormGatherStatsGetWorkspaceSize(totalSum, totalSquareSum, sampleCount, mean, var, momentum, eps, batchMean, batchInvstd, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnSyncBatchNormGatherStatsGetWorkspaceSize 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 aclnnSyncBatchNormGatherStats.
  ret = aclnnSyncBatchNormGatherStats(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnSyncBatchNormGatherStats failed. ERROR: %d\n", ret); return ret);

  // 4. (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);

  // 5. Obtain the output values and copy the results from the device side memory to the host side. Modify this based on the specific API definition.
  auto size = GetShapeSize(batchMeanShape);
  std::vector<float> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), batchMeanDeviceAddr,
                          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. Modify this based on the specific API definition.
  aclDestroyTensor(totalSum);
  aclDestroyTensor(totalSquareSum);
  aclDestroyTensor(sampleCount);
  aclDestroyTensor(mean);
  aclDestroyTensor(var);
  aclDestroyTensor(batchMean);
  aclDestroyTensor(batchInvstd);
  // 7. Release device resources. Modify based on the specific API definition.
  aclrtFree(totalSumDeviceAddr);
  aclrtFree(totalSquareSumDeviceAddr);
  aclrtFree(sampleCountDeviceAddr);
  aclrtFree(meanDeviceAddr);
  aclrtFree(varDeviceAddr);
  aclrtFree(batchMeanDeviceAddr);
  aclrtFree(batchInvstdDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}