aclnnMaxUnpool2dBackward

📄 View Source Code

Applicable Products

Function

• Operator function: backward propagation of the inverse operation of MaxPool2d aclnnMaxUnpool2d, which fills the element values of gradOutput into out according to the indices.

• Formula:

  $$out[n][c][i] = gradOutput[n][c][indices[n][c][i]]$$

  where out, gradOutput, and indices are obtained by reshaping the last two axes into one axis, H and W represent the last two axes respectively, and $i ∈ [0, H*W)$.

Function Prototype

Each operator is divided into a two-phase API. You must first call the "aclnnMaxUnpool2dBackwardGetWorkspaceSize" API to obtain the workspace size required for computation and the executor that contains the operator computation process, and then call the "aclnnMaxUnpool2dBackward" API to perform the computation.

aclnnStatus aclnnMaxUnpool2dBackwardGetWorkspaceSize(
  const aclTensor*     gradOutput,
  const aclTensor*     self,
  const aclTensor*     indices,
  const aclIntArray*   outputSize,
  aclTensor*           out,
  uint64_t*            workspaceSize,
  aclOpExecutor**      executor)

aclnnStatus aclnnMaxUnpool2dBackward(
  void*          workspace,
  uint64_t       workspaceSize,
  aclOpExecutor* executor,
  aclrtStream    stream)

aclnnMaxUnpool2dBackwardGetWorkspaceSize

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-Contiguous Tensor
  gradOutput (aclTensor*)	Input	gradOutput in the formula.	The data type must be convertible to the data type of out (see Conversion Relationship). The data type must be the same as self. The shape must be (N, C, outputSize[0], outputSize[1]) or (C, outputSize[0], outputSize[1]). The number of dimensions must be the same as self.	FLOAT, FLOAT16, INT32, INT64, INT16, INT8, UINT8, BOOL	ND	3-4	√
  self (aclTensor*)	Input	-	Must have the same data type as gradOutput Must have a shape of (N, C, H, W) or (C, H, W), and the shape must be consistent with indices.	FLOAT, FLOAT16, INT32, INT64, INT16, INT8, UINT8, BOOL	ND	3-4	√
  indices (aclTensor*)	Input	The indices in the formula.	The shape must be consistent with self.	INT64, INT32	ND	3-4	√
  outputSize (aclIntArray*)	Input	-	The number of elements must be 2, and the element values must match the last two dimensions of gradOutput's shape.	-	-	-	-
  out (aclTensor*)	Output	The out in the formula.	The shape must be consistent with self.	FLOAT, FLOAT16, INT32, INT64, INT16, INT8, UINT8, BOOL	ND	3-4	√
  workspaceSize (uint64_t*)	Output	Returns the workspace size to be requested 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 gradOutput, self, indices, outputSize, or out is a null pointer.
  ACLNN_ERR_PARAM_INVALID	161002	gradOutput, self, indices, or out is not within the supported data type range.
  		gradOutput cannot be converted to the data type of out.
  		The data types of gradOutput and self are inconsistent.
  		The dimension of self is neither 3D nor 4D.
  		The dimension of gradOutput is inconsistent with that of self.
  		When self is 3D, the sizes of self and gradOutput in the C dimension are inconsistent.
  		When self is 4D, the sizes of self and gradOutput in the N and C dimensions are inconsistent.
  		The shapes of self, indices, and out are inconsistent.
  		The number of elements in outputSize is not 2.
  		The sizes of the last two dimensions of gradOutput do not match the elements of outputSize.
  		self and indices are non-empty tensors, while gradOutput is an empty tensor.

aclnnMaxUnpool2dBackward

• 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 by the first-phase API aclnnLogAddExpGetWorkspaceSize.
  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:

  • aclnnMaxUnpool2dBackward 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_max_unpool2d_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. 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 interface.
  int64_t N = 1;
  int64_t C = 3;
  int64_t H = 2;
  int64_t W = 2;
  std::vector<int64_t> outputSizeData = {3, 1};
  std::vector<int64_t> gradOutputShape = {N, C, outputSizeData[0], outputSizeData[1]};
  std::vector<int64_t> selfShape = {N, C, H, W};
  std::vector<int64_t> indicesShape = {N, C, H, W};
  std::vector<int64_t> outShape = {N, C, H, W};
  void* gradOutputDeviceAddr = nullptr;
  void* selfDeviceAddr = nullptr;
  void* indicesDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* gradOutput = nullptr;
  aclTensor* self = nullptr;
  aclTensor* indices = nullptr;
  aclIntArray* outputSize = nullptr;
  aclTensor* out = nullptr;
  std::vector<float> gradOutputHostData = {0, 1, 2, 3, 4, 5, 6, 7, 8};
  std::vector<float> selfHostData(12, 1);
  std::vector<int32_t> indicesHostData = {0, 1, 2, 1, 2, 0, 1, 1, 1, 0, 0, 0};
  std::vector<float> outHostData(12, 0);
  // 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 indices aclTensor.
  ret = CreateAclTensor(indicesHostData, indicesShape, &indicesDeviceAddr, aclDataType::ACL_INT32, &indices);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the outputSize aclIntArray.
  outputSize = aclCreateIntArray(outputSizeData.data(), 2);
  CHECK_RET(outputSize != nullptr, 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 aclnnMaxUnpool2dBackward.
  ret = aclnnMaxUnpool2dBackwardGetWorkspaceSize(gradOutput, self, indices, outputSize, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMaxUnpool2dBackwardGetWorkspaceSize 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 aclnnMaxUnpool2dBackward.
  ret = aclnnMaxUnpool2dBackward(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMaxUnpool2dBackward 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(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(indices);
  aclDestroyIntArray(outputSize);
  aclDestroyTensor(out);

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