aclnnThreeInterpolateBackward

📄 View source code

Supported Products

Prototype

Each operator has two-phase API calls. First, aclnnThreeInterpolateBackwardGetWorkspaceSize is called to obtain the workspace size required for computation and the executor that contains the operator computation process. Then, aclnnThreeInterpolateBackward is called to perform computation.

• aclnnStatus aclnnThreeInterpolateBackwardGetWorkspaceSize(const aclTensor *grad_x, const aclTensor *idx, const aclTensor *weight, int m, aclTensor *grad_y, uint64_t *workspaceSize, aclOpExecutor **executor)
• aclnnStatus aclnnThreeInterpolateBackward(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream)

Function

• Operator description: Performs three-point interpolation calculation based on grad_x, idx, and weight to obtain the gradient grad_y.

• Formula:

  $$grad\_y[b,c,idx[b,n,i]] = grad\_y[b,c,idx[b,n,i]] + grad\_x[b,c,n]*weight[b,n,i]\\ i\in[0,2]\ b\in[0,B) \ c\in[0,C) \ n\in[0,N)$$

aclnnThreeInterpolateBackwardGetWorkspaceSize

• Parameters:

  • grad_x (aclTensor*, compute input): gradient value of the previous step of network backpropagation. It is an aclTensor on the device. The data type can be FLOAT or FLOAT16. The shape can be (b, c, n). Non-contiguous tensors are supported. The data format can be NCHW.
  • idx (aclTensor*, compute input): three nearest neighbor feature indexes of the target feature. It is an aclTensor on the device. The data type can be INT32 or INT64. The shape can be (b, n, 3). Non-contiguous tensors are supported, but empty tensors are not supported. The data format supports ND.
  • weight (aclTensor*, compute input): three nearest neighbor feature weights of the target feature. It is an aclTensor on the device. The data type can be FLOAT or FLOAT16. The shape can be (b, n, 3). Non-contiguous tensors are supported, but empty tensors are not supported. The data format supports ND.
  • m (uint32_t*, compute input): size of the lowest dimension of the input feature. It derives the shape of grad_y for infershape. The data type can be INT32.
  • grad_y (aclTensor*, compute output): gradient calculation result. It is an aclTensor on the device. The data type can be FLOAT or FLOAT16. The shape can be (b, c, m). Non-contiguous tensors are supported, but empty tensors are not supported. The data format supports NCHW.
  • workspaceSize (uint64_t*, output): size of the workspace to be allocated on the device.
  • executor (aclOpExecutor**, output): operator executor, containing the operator computation process.

• Returns:

  aclnnStatus: status code. For details, see aclnn Return Codes.

The first-phase API implements input parameter verification. The following errors may be thrown.
161001 (ACLNN_ERR_PARAM_NULLPTR): 1. The passed grad_x, idx, weight, or grad_y is a null pointer.
161002 (ACLNN_ERR_PARAM_INVALID): 1. The data type or data format of grad_x, idx, weight, m, or grad_y is not supported.
                                  2. When grad_x, idx, weight, and grad_y are not null pointers, their data types, formats, or shapes are not supported.

aclnnThreeInterpolateBackward

• Parameters:

  • workspace (void*, input): address of the workspace to be allocated on the device.
  • workspaceSize (uint64_t, input): size of the workspace to be allocated on the device, which is obtained by calling the first-phase API aclnnThreeInterpolateBackwardGetWorkspaceSize.
  • executor (aclOpExecutor*, input): operator executor, containing the operator computation process.
  • stream (aclrtStream, input): stream for executing the task.

• Returns:

  aclnnStatus: status code. For details, see aclnn Return Codes.

Constraints

• The values in idx must be less than m.
• Deterministic computing:
  • aclnnThreeInterpolateBackward defaults to a non-deterministic implementation. Enabling deterministic computing is not supported.

Example

The following example is for reference only. For details, see Compilation and Running Sample.

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnn/aclnn_base.h"

#ifdef __cplusplus
extern "C" {
#endif

/**
 * @brief Second-phase API of aclnnThreeInterpolateBackward, which is used to perform computation.
*/
aclnnStatus aclnnThreeInterpolateBackward(void *workspace, uint64_t workspaceSize,
                                          aclOpExecutor *executor, aclrtStream stream);

/**
 * @brief First-phase API of aclnnThreeInterpolateBackward, which is used to calculate the workspace size based on the specific computation process.
 * @domain aclnn_ops_train
*/
aclnnStatus aclnnThreeInterpolateBackwardGetWorkspaceSize(const aclTensor *grad_x, const aclTensor *idx,
                                                          const aclTensor *weight, int m, aclTensor *grad_y,
                                                          uint64_t *workspaceSize, aclOpExecutor **executor);
#ifdef __cplusplus
}
#endif

#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;
}

void PrintOutResult(std::vector<int64_t> &shape, void** deviceAddr) {
  auto size = GetShapeSize(shape);
  std::vector<float> resultData(size, 0);
  auto ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]),
                         *deviceAddr, 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);
  for (int64_t i = 0; i < size; i++) {
    LOG_PRINT("result[%ld] is: %f\n", i, resultData[i]);
  }
}

int Init(int32_t deviceId, aclrtStream* stream) {
  // (Fixed writing) Initialize resources.
  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, aclFormat format) {
  auto size = GetShapeSize(shape) * sizeof(T);
  // Call aclrtMalloc to allocate memory on the device.
  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);
  // Call aclrtMemcpy to copy the data on the host to the memory on the device.
  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 the 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 aclCreateTensor to create an aclTensor.
  *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, format,
                            shape.data(), shape.size(), *deviceAddr);
  return 0;
}

int main() {
  // 1. (Fixed writing) Initialize the device and stream. For details, see the ACL API manual.
  // Set the device ID in use.
  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);

  uint32_t bs = 1;
  uint32_t cs = 2;
  uint32_t ms = 6;
  uint32_t ns = 5;
  uint32_t c0 = 16;

  auto c1 = (cs + c0 - 1) / c0;

  // 2. Construct the input and output based on the API.
  std::vector<int64_t> gradXShape = {bs, cs, ns, 1}; // nchw
  std::vector<int64_t> idxShape = {bs, ns, 3}; // nd
  std::vector<int64_t> weightShape = {bs, ns, 3}; // nd
  std::vector<int64_t> gradYShape = {bs, cs, ms, 1}; // nchw

  void* gradXDeviceAddr = nullptr;
  void* idxDeviceAddr = nullptr;
  void* weightDeviceAddr = nullptr;
  void* gradYDeviceAddr = nullptr;

  aclTensor* gradX = nullptr;
  aclTensor* idx = nullptr;
  aclTensor* weight = nullptr;
  aclTensor* gradY = nullptr;

  std::vector<float> gradXHostData = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
  std::vector<uint32_t> idxHostData = {0, 3, 2, 3, 5, 1, 0, 3, 0, 0, 3, 5, 0, 4, 1};
  std::vector<float> weightHostData = {3.2081969,-0.86573875,-1.0929844,5.6045847,5.328887,-2.0821328,0.98572457,9.612394,2.216394,
                                       1.8722068,4.2976365,-7.019285,-8.4070425,1.0159919,-0.78814566};
  std::vector<float> gradYHostData = {0};

  ret = CreateAclTensor(gradXHostData, gradXShape, &gradXDeviceAddr, aclDataType::ACL_FLOAT, &gradX, aclFormat::ACL_FORMAT_NCHW);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(idxHostData, idxShape, &idxDeviceAddr, aclDataType::ACL_INT32, &idx, aclFormat::ACL_FORMAT_ND);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(weightHostData, weightShape, &weightDeviceAddr, aclDataType::ACL_FLOAT, &weight, aclFormat::ACL_FORMAT_ND);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(gradYHostData, gradYShape, &gradYDeviceAddr, aclDataType::ACL_FLOAT, &gradY, aclFormat::ACL_FORMAT_NCHW);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  // 3. Call the CANN operator library API, which needs to be replaced with the actual API.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;

  // Call the first-phase API of aclnnThreeInterpolateBackward.
  ret = aclnnThreeInterpolateBackwardGetWorkspaceSize(gradX, idx, weight, ms, gradY, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnThreeInterpolateBackwardGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);

  // Allocate device memory based on the computed workspaceSize.
  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 aclnnThreeInterpolateBackward.
  ret = aclnnThreeInterpolateBackward(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnThreeInterpolateBackward failed. ERROR: %d\n", ret); return ret);

  // 4. (Fixed writing) Wait until the task execution is 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 to the host. Modify the code based on the API definition.
  PrintOutResult(gradYShape, &gradYDeviceAddr);

  // 6. Release aclTensor and aclScalar. Modify the configuration based on the API definition.
  aclDestroyTensor(gradX);
  aclDestroyTensor(idx);
  aclDestroyTensor(weight);
  aclDestroyTensor(gradY);

  // 7. Release device resources. Modify the configuration based on the API definition.
  aclrtFree(gradXDeviceAddr);
  aclrtFree(idxDeviceAddr);
  aclrtFree(weightDeviceAddr);
  aclrtFree(gradYDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();
  return 0;
}