aclnnGather

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: Gathers data along a specified dimension dim of the input tensor.
Formula:

Given a tensor $self$ , a dimension $d$ , and an index tensor $index$ , let $n$ be the number of dimensions of $self$ , $i_d$ denote the index along dimension $d$ , and $index_{i_d}$ denote the $i_d$ -th index value of the index tensor $index$ along dimension $d$ . The gather operation along the specified dimension $d$ can be expressed by the following mathematical formula:
$gather(X,index,d)_{i_0,i_1,\cdots,i_{d-1},i_{d+1},\cdots,i_{n-1}} = self_{i_0,i_1,\cdots,i_{d-1},index_{i_d},i_{d+1},\cdots,i_{n-1}}$
Example:
- Example 1:
  
  Assume the input tensor $self=\begin{bmatrix}1 & 2 & 3\\ 4 & 5 & 6\\ 7 & 8 & 9\end{bmatrix}$ , the index tensor $index=\begin{bmatrix}0 & 2\\ 1 & 0\end{bmatrix}$ , and $dim = 0$ . Then the output tensor $out=\begin{bmatrix}1 & 8\\ 4 & 2\end{bmatrix}$ . The specific computation process is as follows:
  $\begin{aligned} out_{0,0}&=self_{index_{0,0}, 0}=self_{0,0}=1 \\ out_{0,1}&=self_{index_{0,1}, 1}=self_{2,1}=8 \\ out_{1,0}&=self_{index_{1,0}, 0}=self_{1,0}=4 \\ out_{1,1}&=self_{index_{1,1}, 1}=self_{0,1}=2 \end{aligned}$
- Example 2:
  
  Assume the input tensor $self=\begin{bmatrix}1 & 2 & 3\\ 4 & 5 & 6\\ 7 & 8 & 9\end{bmatrix}$ , the index tensor $index=\begin{bmatrix}0 & 2\\ 1 & 0\end{bmatrix}$ , and $dim = 1$ . Then the output tensor $out=\begin{bmatrix}1 & 3\\ 5 & 4\end{bmatrix}$ . The specific computation process is as follows:
  $\begin{aligned} out_{0,0}&=self_{0, index_{0,0}}=self_{0,0}=1 \\ out_{0,1}&=self_{0, index_{0,1}}=self_{0,2}=3 \\ out_{1,0}&=self_{1, index_{1,0}}=self_{1,1}=5 \\ out_{1,1}&=self_{1, index_{1,1}}=self_{1,0}=4 \end{aligned}$

Function Prototype

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

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

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

aclnnGatherGetWorkspaceSize

Parameters

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
self (aclTensor*)	Input	self in the formula.	Must have the same data type as out and the same number of dimensions as index.	DOUBLE, FLOAT16, BFLOAT16, FLOAT32, INT32, UINT32, INT64, UINT64, INT16, UINT16, INT8, UINT8, BOOL	-	0-8	√
dim (int64_t)	Input	d in the formula.	Value range: [-self.dim(), self.dim() - 1]	-	-	-	-
index (aclTensor*)	Input	The index in the formula.	The number of dimensions must be the same as that of self, and the shape must be the same as that of out. Except for the dimension specified by dim, the size of other dimensions must be less than or equal to the size of the corresponding dimension of self. The specific values in index represent the subscripts on the corresponding dim axis of self, with a value range of [-self.shape[dim], self.shape[dim] - 1]. Out-of-bounds indexing is not supported in index.	INT32, INT64	-	0-8	√
out (aclTensor*)	Output	Output aclTensor*.	The data type must be the same as self, and the shape must be the same as index.	Same as self.	-	Same as index.	√
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 Value	Error Code	Description
ACLNN_ERR_PARAM_NULLPTR	161001	The input self, index, or out is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data type of self, index, or out is not within the supported range.
		The data types of out and self are inconsistent.
		The dimensions of self and index are different.
		The dimension of self, index, or out exceeds 8.
		The shapes of out and index are inconsistent.
		The dimension specified by dim exceeds the valid range of self dimensions [-self.dim(), self.dim() - 1].
		The size of index in dimensions other than the one specified by dim is greater than that of self.
		Index is a non-empty tensor and the size of self in the dimension specified by dim is 0.

aclnnGather

Parameters

Parameter	Input/Output	Description
workspace	Input	Specifies the workspace memory address applied on the Device side.
workspaceSize	Input	Specifies the workspace size applied on the Device side, which is obtained through the first-phase API aclnnGatherGetWorkspaceSize.
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:
- aclnnGather 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.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 input and output. Customize based on the API interface.
  std::vector<int64_t> selfShape = {4, 2};
  std::vector<int64_t> indexShape = {4, 2};
  std::vector<int64_t> outShape = {4, 2};
  void* selfDeviceAddr = nullptr;
  void* indexDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* self = nullptr;
  aclTensor* index = nullptr;
  aclTensor* out = nullptr;
  std::vector<float> selfHostData = {0, 1, 2, 3, 4, 5, 6, 7};
  std::vector<int64_t> indexHostData = {0, 1, 2, 3, 3, 2, 1, 0};
  std::vector<float> outHostData(8, 0);
  int64_t dim = 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. Modify to the specific API name.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // Call the first-phase API of aclnnGather.
  ret = aclnnGatherGetWorkspaceSize(self, dim, index, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGatherGetWorkspaceSize 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 aclnnGather.
  ret = aclnnGather(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGather 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 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. Modify based on the specific API definition.
  aclDestroyTensor(self);
  aclDestroyTensor(index);
  aclDestroyTensor(out);

  // 7. Release device resources.
  aclrtFree(selfDeviceAddr);
  aclrtFree(indexDeviceAddr);
  aclrtFree(outDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}