aclnnGatherV3

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: Extracts elements from the input tensor along the specified dimension dim based on the index values in index, where batchDims represents the number of batch dimensions. The result is stored in the out tensor.
Example:

For example, when batchDims is 0, given the input tensor $self=\begin{bmatrix}1 & 2 & 3 \\ 4 & 5 & 6 \\ 7 & 8 & 9\end{bmatrix}$ and the index tensor index=[1, 0],.
- Result when dim=0: $out=\begin{bmatrix}4 & 5 & 6 \\ 1 & 2 & 3\end{bmatrix}$
- Result when dim=1: $out=\begin{bmatrix}2 & 1\\ 5 & 4\\ 8 & 7\end{bmatrix}$
The specific computation process is as follows: Taking a 3D tensor as an example, a tensor self with shape (3,2,2) = $\begin{bmatrix}[[1,&2],&[3,&4]], \\ [[5,&6],&[7,&8]], \\ [[9,&10],&[11,&12]]\end{bmatrix}$ , index = [1, 0], the subscripts corresponding to dim=0, 1, 2 of the self tensor are $l$ , $m$ , $n$ respectively, and index is one-dimensional (for the zero-dimensional case: it is treated as a one-dimensional tensor of size 1).
- When dim is 0: I = index[i]; out $[i][m][n]$ = self $[I][m][n]$
- When dim is 1: J = index[j]; out $[l][j][n]$ = self $[l][J][n]$
- When dim is 2: K = index[k]; out $[l][m][k]$ = self $[l][m][K]$
When batchDims is 1: Taking a four-dimensional tensor as an example, for a tensor self with shape (3,3,2,2) and a tensor index with shape (3,2), this is equivalent to performing three gather operations with batchDims set to 0 and dim = dim - batchDims.

Function Prototype

Each operator is divided into a two-phase API. The "aclnnGatherV3GetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnGatherV3" API is called to perform the computation.

aclnnStatus aclnnGatherV3GetWorkspaceSize(
 const aclTensor *self,
 int64_t          dim,
 const aclTensor *index,
 int64_t          batchDims,
 int64_t          mode,
 aclTensor       *out,
 uint64_t        *workspaceSize,
 aclOpExecutor  **executor)

aclnnStatus aclnnGatherV3(
 void          *workspace,
 uint64_t       workspaceSize,
 aclOpExecutor *executor,
 aclrtStream    stream)

aclnnGatherV3GetWorkspaceSize

Parameters

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
self	Input	Data to be gathered.	-	FLOAT, FLOAT16, INT64, INT32, INT16, INT8, UINT8, BOOL, DOUBLE, COMPLEX64	ND	1-8	√
dim	Input	The axis along which to gather.	The value range is [-self.dim(), self.dim()-1].	INT64	-	-	-
index	Input	Index for gathering data.	The value range is [0, self.shape[dim]) (inclusive of 0, exclusive of self.shape[dim]). When batchDims = N and N != 0, the first N dimensions of index are consistent with self.	INT64, INT32	ND	1-8	√
batchDims	Input	Batch dimension for the operation.	The value range is [0, dim], and must be less than or equal to rank(index).	INT64	-	-	-
mode	Input	Computation mode.	Value range: [0, 2]. 0: Performance optimization for index scattering scenarios. 1: Performance optimization for index gathering scenarios. 2: Supports out-of-bounds indexing. Currently only 1 is supported.	INT64	-	-	-
out	Output	Output aclTensor.	When batchDims = N and N == 0, the number of dimensions equals the sum of the number of dimensions of self and index minus one. All dimensions except dim are the same as the corresponding dimensions of self, while the dim dimension is expanded to match the shape of index. When N != 0, the dimensions of output are self[:dim] + index[N+1:] + self[dim+1:].	Same as self.	ND	1-8	-
workspaceSize	Output	Returns the workspace size to be requested on the Device side.	-	-	-	-	-
executor	Output	Returns the op executor, which contains the operator computation process.	-	-	-	-	-

Atlas A2 training products/Atlas A2 inference products, Atlas A3 training products/Atlas A3 inference products: dim currently supports only 0, and batchDims currently supports only 0.

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 self, index, and out parameters are null pointers.
ACLNN_ERR_PARAM_INVALID	161002	The data type of self or index is not within the supported range.
		The number of dimensions of self or index is greater than 8.
		The data type of self and out is inconsistent.
		The shape of out must satisfy that, except for dimension 0 which is expanded to match the shape of index, the lengths of all other dimensions are consistent with the corresponding dimensions of self.

aclnnGatherV3

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 aclnnGatherV3GetWorkspaceSize.
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:
- aclnnGatherV3 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_gather_v3.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);
  // 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 aclCreateTensor 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. Refer to the ACL API manual.
  // Fill in the deviceId based on your 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 = {2, 2};
  std::vector<int64_t> indexShape = {2, 2};
  std::vector<int64_t> outShape = {2, 2, 2};
  void* selfDeviceAddr = nullptr;
  void* indexDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* self = nullptr;
  aclTensor* index = nullptr;
  aclTensor* out = nullptr;
  int64_t dim = 0;
  int64_t batchDims = 0;
  int64_t mode = 1;
  std::vector<float> selfHostData = {0, 1, 2, 3};
  std::vector<int64_t> indexHostData = {1, 0, 0, 1};
  std::vector<float> outHostData = {0, 0, 0, 0, 0, 0, 0, 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 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 aclnnGatherV3.
  ret = aclnnGatherV3GetWorkspaceSize(self, dim, index, batchDims, mode, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGatherV3GetWorkspaceSize 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 aclnnGatherV3.
  ret = aclnnGatherV3(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGatherV3 failed. ERROR: %d\n", ret); return ret);

  // 4. (Boilerplate) Synchronously wait for the task 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 memory to the host. 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(out);

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