aclnnForeachRoundOffNumber

📄 View Source Code

Applicable Products

Function

• Description: Performs a rounding operation with a specified precision on each tensor in the input tensor list. The rounding mode can be specified through roundMode.

• Formula:

  $$x = [{x_0}, {x_1}, ... {x_{n-1}}]\\ y = [{y_0}, {y_1}, ... {y_{n-1}}]\\$$ $$y_i = round(x_i, roundMode) (i=0,1,...n-1)$$

Function Prototype

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

aclnnStatus aclnnForeachRoundOffNumberGetWorkspaceSize(
  const aclTensorList *x,
  const aclTensor     *roundMode,
  const aclTensorList *out,
  uint64_t            *workspaceSize,
  aclOpExecutor      **executor)

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

aclnnForeachRoundOffNumberGetWorkspaceSize

• Parameters

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  x (aclTensorList*)	Input	Represents the list of input tensors for the rounding operation, corresponding to `x` in the formula.	Supports empty tensors. All tensors in this parameter must have the same data type.	FLOAT32, FLOAT16, BFLOAT16	ND	0-8	√
  roundMode (aclTensor*)	Input	Represents the input tensor for the rounding computation. Corresponds to `roundMode` in the formula.	Empty tensors are not supported. The number of elements is 1. The values of roundMode and their corresponding rounding strategies are as follows: When roundMode=1, a round-half-to-even rounding operation is performed on the input. When roundMode=2, a round-toward-negative-infinity rounding operation is performed on the input. When roundMode=3, a round-toward-positive-infinity rounding operation is performed on the input. When roundMode=4, a round-half-away-from-zero rounding operation is performed on the input. When roundMode=5, a round-toward-zero rounding operation is performed on the input. When roundMode=6, a round-to-nearest-odd rounding operation is performed on the input. For other roundMode values, a round-half-to-even rounding operation is performed if precision loss occurs; otherwise, no rounding operation is performed.	INT8	ND	0-8	√
  out (aclTensorList*)	Output	Represents the output tensor list for the rounding operation. Corresponds to `y` in the formula.	Supports empty tensors. All tensors in this parameter must have the same data type. The data type and data format are consistent with those of the input `x`, and the shape size is greater than or equal to the shape size of the input `x`.	FLOAT32, FLOAT16, BFLOAT16	ND	0-8	√
  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 Code	Error Code	Description
  ACLNN_ERR_PARAM_NULLPTR	161001	Indicates that x, roundMode, or out is a null pointer.
  ACLNN_ERR_PARAM_INVALID	161002	The data types of x, roundMode, and out are not within the supported range.
  		The data types of x and out are inconsistent.
  ACLNN_ERR_INNER_TILING_ERROR	561002	The shapes of x and out do not satisfy the constraints.
  		The data types of tensors in x or out are inconsistent.
  		The dimension of a tensor in x or out exceeds 8.
  		The number of elements in roundMode is not 1.

aclnnForeachRoundOffNumber

• 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 aclnnForeachRoundOffNumberGetWorkspaceSize.
  executor	Input	Operator executor, which 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:

  • aclnnForeachRoundOffNumber 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_foreach_round_off_number.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 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 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 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. 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> selfShape1 = {2, 3};
  std::vector<int64_t> selfShape2 = {1, 3};
  std::vector<int64_t> outShape1 = {2, 3};
  std::vector<int64_t> outShape2 = {1, 3};
  std::vector<int64_t> alphaShape = {1};
  void* input1DeviceAddr = nullptr;
  void* input2DeviceAddr = nullptr;
  void* out1DeviceAddr = nullptr;
  void* out2DeviceAddr = nullptr;
  void* alphaDeviceAddr = nullptr;
  aclTensor* input1 = nullptr;
  aclTensor* input2 = nullptr;
  aclTensor* alpha = nullptr;
  aclTensor* out1 = nullptr;
  aclTensor* out2 = nullptr;
  std::vector<float> input1HostData = {1.1, 2.2, 3.3, 4.5, 5.3, 6.12};
  std::vector<float> input2HostData = {7.45, 8.87, 9.54};
  std::vector<float> out1HostData(6, 0);
  std::vector<float> out2HostData(3, 0);
  std::vector<float> alphaHostData = {1};
  // Create the input1 aclTensor.
  ret = CreateAclTensor(input1HostData, selfShape1, &input1DeviceAddr, aclDataType::ACL_FLOAT, &input1);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the input2 aclTensor.
  ret = CreateAclTensor(input2HostData, selfShape2, &input2DeviceAddr, aclDataType::ACL_FLOAT, &input2);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the alpha aclTensor.
  ret = CreateAclTensor(alphaHostData, alphaShape, &alphaDeviceAddr, aclDataType::ACL_INT8, &alpha);
  CHECK_RET(alpha != nullptr, return ret);
  // Create the out1 aclTensor.
  ret = CreateAclTensor(out1HostData, outShape1, &out1DeviceAddr, aclDataType::ACL_FLOAT, &out1);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create out2 aclTensor.
  ret = CreateAclTensor(out2HostData, outShape2, &out2DeviceAddr, aclDataType::ACL_FLOAT, &out2);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  std::vector<aclTensor*> tempInput{input1, input2};
  aclTensorList* tensorListInput = aclCreateTensorList(tempInput.data(), tempInput.size());
  std::vector<aclTensor*> tempOutput{out1, out2};
  aclTensorList* tensorListOutput = aclCreateTensorList(tempOutput.data(), tempOutput.size());

  // 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 aclnnForeachRoundOffNumber.
  ret = aclnnForeachRoundOffNumberGetWorkspaceSize(tensorListInput, alpha, tensorListOutput, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnForeachRoundOffNumberGetWorkspaceSize 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 aclnnForeachRoundOffNumber.
  ret = aclnnForeachRoundOffNumber(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnForeachRoundOffNumber failed. ERROR: %d\n", ret); return ret);

  // 4. (Boilerplate) Synchronously wait for task execution to finish.
  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(outShape1);
  std::vector<float> out1Data(size, 0);
  ret = aclrtMemcpy(out1Data.data(), out1Data.size() * sizeof(out1Data[0]), out1DeviceAddr,
                    size * sizeof(out1Data[0]), ACL_MEMCPY_DEVICE_TO_HOST);
  for (int64_t i = 0; i < size; i++) {
    LOG_PRINT("out1 result[%ld] is: %f\n", i, out1Data[i]);
  }

  size = GetShapeSize(outShape2);
  std::vector<float> out2Data(size, 0);
  ret = aclrtMemcpy(out2Data.data(), out2Data.size() * sizeof(out2Data[0]), out2DeviceAddr,
                    size * sizeof(out2Data[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("out2 result[%ld] is: %f\n", i, out2Data[i]);
  }

  // 6. Release aclTensor. Modify based on the specific API definition.
  aclDestroyTensorList(tensorListInput);
  aclDestroyTensorList(tensorListOutput);
  aclDestroyTensor(alpha);

  // 7. Release device resources. Modify based on the specific API definition.
  aclrtFree(input1DeviceAddr);
  aclrtFree(input2DeviceAddr);
  aclrtFree(out1DeviceAddr);
  aclrtFree(out2DeviceAddr);
  aclrtFree(alphaDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();
  return 0;
}