aclnnScatterAdd

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: Fills values from the src tensor into the self tensor element by element according to the specified dimension and the positional relationships in the index tensor. If more than one src value is filled into the same position of self, these values are accumulated at that position.

Example: For a 3D tensor, self is updated according to the following rules:
```
self[index[i][j][k]][j][k] += src[i][j][k] # If dim == 0
self[i][index[i][j][k]][k] += src[i][j][k] # If dim == 1
self[i][j][index[i][j][k]] += src[i][j][k] # If dim == 2
```

The following requirements must be met during computation:

self, index, and src must have the same number of dimensions.
- For each dimension d, index.size(d) <= src.size(d).
- For each dimension d where d != dim, index.size(d) <= self.size(d).
- dim ranges from [-self.dim(), self.dim() - 1].
Example:

Input tensor $self = \begin{bmatrix} [1&2&3] \\ [4&5&6] \\ [7&8&9] \end{bmatrix}$ , index tensor $index = \begin{bmatrix} [0&2&1] \\ [0&0&1] \end{bmatrix}$ , dim = 1, source tensor $src = \begin{bmatrix} [10&11&12] \\ [13&14&15] \end{bmatrix}$ , output tensor $output = \begin{bmatrix} [11&14&14] \\ [31&20&6] \\ [7&8&9] \end{bmatrix}$

dim = 1 indicates that scatter_add accumulates along the columns of the tensor based on $index$ .

$output[0][0] = self[0][0] + src[0][0]$ = 1 + 10,

$output[0][1] = self[0][1] + src[0][2]$ = 2 + 12,

$output[0][2] = self[0][2] + src[0][1]$ = 3 + 11,

$output[1][0] = self[1][0] + src[1][0] + src[1][1]$ = 4 + 13 + 14,

$output[1][1] = self[1][1] + src[1][2]$ = 5 + 15,

$output[1][2] = self[1][2]$ = 6,

$output[2][0] = self[2][0]$ = 7,

$output[2][1] = self[2][1]$ = 8,

$output[2][2] = self[2][2]$ = 9.

Here, the Number of Dimensions of $self$ , $index$ , and $src$ are all 2. Each dimension size {2, 3} of $index$ is not greater than the corresponding dimension size {2, 3} of $src$ . In the dimension where dim != 1 (dim = 0), the dimension size {2} of $index$ is not greater than the corresponding dimension size {3} of $self$ . The maximum value {2} in $index$ is less than the size {3} of $self$ in the dim = 1 dimension.

Function Prototype

Each operator is divided into a two-phase API. The "aclnnScatterAddGetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnScatterAdd" API is called to perform the computation.

aclnnStatus aclnnScatterAddGetWorkspaceSize(
  const aclTensor* self,
  int64_t          dim,
  const aclTensor* index,
  const aclTensor* src,
  aclTensor*       out,
  uint64_t*        workspaceSize,
  aclOpExecutor**  executor)

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

aclnnScatterAddGetWorkspaceSize

Parameters:

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Shape	Non-contiguous Tensor
self (aclTensor*)	Input	The input `self` in the formula, which is the target tensor for scatter.	The number of dimensions must be the same as that of index and src. The data type must be consistent with that of src.	BFLOAT16, FLOAT16, FLOAT32, DOUBLE, INT64, INT32, INT16, INT8, UINT8, BOOL, COMPLEX64, COMPLEX128	ND	0-8	√
dim (int64_t)	Input	The input `dim` in the formula.	-	-	-	-	-
index (aclTensor*)	Input	Input `index` in the formula.	The number of dimensions must be the same as that of src.	INT32, INT64	ND	-	√
src (aclTensor*)	Input	Input `src` in the formula, the source tensor.	Number of Dimensions must be the same as that of index. Data type must be consistent with that of self.	BFLOAT16, FLOAT16, FLOAT32, DOUBLE, INT64, INT32, INT16, INT8, UINT8, BOOL, COMPLEX64, COMPLEX128	ND	-	×
out (aclTensor*)	Output	The `output` in the formula.	The shape must be the same as that of self. The data type is the same as that of self.	BFLOAT16, FLOAT16, FLOAT32, DOUBLE, INT64, INT32, INT16, INT8, UINT8, BOOL, COMPLEX64, COMPLEX128	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	Return Code	Description
ACLNN_ERR_PARAM_NULLPTR	161001	The passed self, index, src, or out is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data type of self, index, src, or out is not within the supported range.
		The shapes of self and out are inconsistent.
		The shape of src or index is invalid.

aclnnScatterAdd

Parameters:

Parameter	Input/Output	Description
workspace	Input	Address of the workspace memory applied on the Device side.
workspaceSize	Input	Size of the workspace memory applied on the Device side, obtained from the first-phase API aclnnScatterAddGetWorkspaceSize.
executor	Input	Operator executor, which contains the operator computation flow.
stream	Input	Specifies the stream that executes the task.

Return value

aclnnStatus: return code. For details, see aclnn Return Code.

Constraints

Deterministic computation:
- aclnnScatterAdd 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_scatter_add.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. 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) {
  auto size = GetShapeSize(shape) * sizeof(T);
  // Call aclrtMalloc to apply for device 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);
  // Call aclrtMemcpy to copy data from the host to the device memory.
  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> selfShape = {4, 4};
  std::vector<int64_t> indexShape = {3, 4};
  std::vector<int64_t> srcShape = {4, 4};
  std::vector<int64_t> outShape = {4, 4};
  int64_t dim = 0;
  void* selfDeviceAddr = nullptr;
  void* indexDeviceAddr = nullptr;
  void* srcDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* self = nullptr;
  aclTensor* index = nullptr;
  aclTensor* src = nullptr;
  aclTensor* out = nullptr;
  std::vector<float> selfHostData(16, 0);
  std::vector<int64_t> indexHostData = {0, 1, 2, 1, 0, 1, 2, 0, 2, 2, 1, 0};
  std::vector<float> srcHostData = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15};
  std::vector<float> outHostData(16, 0);
  // Create the self aclTensor.
  ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the index aclTensor.
  ret = CreateAclTensor(indexHostData, indexShape, &indexDeviceAddr, aclDataType::ACL_INT64, &index);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the src aclTensor.
  ret = CreateAclTensor(srcHostData, srcShape, &srcDeviceAddr, aclDataType::ACL_FLOAT, &src);
  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 it with the specific API name.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // Call the first-phase API of aclnnScatterAdd.
  ret = aclnnScatterAddGetWorkspaceSize(self, dim, index, src, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnScatterAddGetWorkspaceSize 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 aclnnScatterAdd.
  ret = aclnnScatterAdd(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnScatterAdd 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 side memory to the host side. 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(self);
  aclDestroyTensor(index);
  aclDestroyTensor(src);
  aclDestroyTensor(out);
  // 7. Release device resources. Modify parameters based on the specific API definition.
  aclrtFree(selfDeviceAddr);
  aclrtFree(indexDeviceAddr);
  aclrtFree(srcDeviceAddr);
  aclrtFree(outDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();
  return 0;
}