aclnnSort

接口原型

每个算子有两段接口，必须先调用“aclnnXxxGetWorkspaceSize”接口获取入参并根据计算流程计算所需workspace大小，再调用“aclnnXxx”接口执行计算。两段式接口如下：

第一段接口：aclnnStatus aclnnSortGetWorkspaceSize(const aclTensor *self, bool stable, int64_t dim, bool descending, aclTensor *valuesOut, aclTensor *indicesOut, uint64_t *workspaceSize, aclOpExecutor **executor)
第二段接口：aclnnStatus aclnnSort(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream)

功能描述

算子功能：对张量中的元素根据某个维度进行升序/降序排序，并返回对应的下标index值。例如输入张量是N维[0, N-1]，需要根据维度[k]进行排序。

aclnnSortGetWorkspaceSize

接口定义：
aclnnStatus aclnnSortGetWorkspaceSize(const aclTensor *self, bool stable, int64_t dim, bool descending, aclTensor *valuesOut, aclTensor *indicesOut, uint64_t *workspaceSize, aclOpExecutor **executor)
参数说明：
- self： Device侧的aclTensor，数据类型支持FLOAT16、FLOAT32、INT8、INT16、INT32、INT64、UINT8。支持非连续的Tensor。数据格式支持ND。(当输入是FLOAT32时，Atlas 训练系列产品会将其转换成FLOAT16进行排序，然后再转换回FLOAT32；Atlas A2训练系列产品会保持FLOAT32进行排序。)
- stable：是否稳定排序，数据类型为BOOL。True为稳定排序，False为非稳定排序。
- dim：排序标准的维度，数据类型支持INT64。取值范围为 [-self.dim(), self.dim()-1]。
- descending：控制排序顺序，数据类型为BOOL。True为降序，False为升序。
- valuesOut：Device侧的aclTensor，数据类型支持FLOAT16、FLOAT32、DOUBLE、INT8、INT16、INT32、INT64、UINT8。支持非连续的Tensor。数据格式支持ND。shape需要与self一致。
- indicesOut：Device侧的aclTensor，数据类型支持INT64。支持非连续的Tensor，数据格式支持ND。shape需要与self一致。
- workspaceSize：返回用户需要在Device侧申请的workspace大小。
- executor：返回op执行器，包含了算子计算流程。
返回值：
返回aclnnStatus状态码，具体参见aclnn返回码。
第一段接口完成入参校验，出现以下场景时报错：
- 返回161001（ACLNN_ERR_PARAM_NULLPTR）：传入的self、valuesOut或indicesOut是空指针。
- 返回161002（ACLNN_ERR_PARAM_INVALID）：
  - self、valuesOut或indicesOut的数据类型或数据格式不在支持的范围内，或者shape相互不匹配。
  - dim的取值不在支持的范围内。

aclnnSort

接口定义：
aclnnStatus aclnnSort(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream)
参数说明：
- workspace：在Device侧申请的workspace内存起址。
- workspaceSize：在Device侧申请的workspace大小，由第一段接口aclnnSortGetWorkspaceSize获取。
- executor：op执行器，包含了算子计算流程。
- stream：指定执行任务的AscendCL stream流。
返回值：
返回aclnnStatus状态码，具体参见aclnn返回码。

调用示例

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_sort.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, aclrtContext* context, aclrtStream* stream) {
  // 固定写法，AscendCL初始化
  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 = aclrtCreateContext(context, deviceId);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateContext failed. ERROR: %d\n", ret); return ret);
  ret = aclrtSetCurrentContext(*context);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetCurrentContext 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);
  // 调用aclrtMalloc申请device侧内存
  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);
  // 调用aclrtMemcpy将Host侧数据拷贝到device侧内存上
  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);

  // 计算连续tensor的strides
  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];
  }

  // 调用aclCreateTensor接口创建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. （固定写法）device/context/stream初始化，参考AscendCL对外接口列表
  // 根据自己的实际device填写deviceId
  int32_t deviceId = 0;
  aclrtContext context;
  aclrtStream stream;
  auto ret = Init(deviceId, &context, &stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);

  // 2. 构造输入与输出，需要根据API的接口自定义构造
  bool stable = false;
  int64_t dim = 0;
  bool descending = false;
  std::vector<int64_t> selfShape = {3, 4};
  std::vector<int64_t> outValuesShape = {3, 4};
  std::vector<int64_t> outIndicesShape = {3, 4};
  void* selfDeviceAddr = nullptr;
  void* outValuesDeviceAddr = nullptr;
  void* outIndicesDeviceAddr = nullptr;
  aclTensor* self = nullptr;
  aclTensor* outValues = nullptr;
  aclTensor* outIndices = nullptr;
  std::vector<int64_t> selfHostData = {7, 8, 9, 10, 11, 12, 1, 2, 3, 4, 5, 6};
  std::vector<int64_t> outValuesHostData = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
  std::vector<int64_t> outIndicesHostData = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
  // 创建self aclTensor
  ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_INT64, &self);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 创建outValues和outIndices aclTensor
  ret = CreateAclTensor(outValuesHostData, outValuesShape, &outValuesDeviceAddr, aclDataType::ACL_INT64, &outValues);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(outIndicesHostData, outIndicesShape, &outIndicesDeviceAddr, aclDataType::ACL_INT64, &outIndices);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  // 3. 调用CANN算子库API，需要修改为具体的API名称
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // 调用aclnnSort第一段接口
  ret = aclnnSortGetWorkspaceSize(self, stable, dim, descending, outValues, outIndices, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnSortGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
  // 根据第一段接口计算出的workspaceSize申请device内存
  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);
  }
  // 调用aclnnSort第二段接口
  ret = aclnnSort(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnSort failed. ERROR: %d\n", ret); return ret);

  // 4. （固定写法）同步等待任务执行结束
  ret = aclrtSynchronizeStream(stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);

  // 5. 获取输出的值，将device侧内存上的结果拷贝至Host侧，需要根据具体API的接口定义修改
  auto size = GetShapeSize(outValuesShape);
  std::vector<int64_t> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), outValuesDeviceAddr,
                    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 values [%ld] is: %ld\n", i, resultData[i]);
  }

  auto size2 = GetShapeSize(outIndicesShape);
  std::vector<int64_t> resultData2(size2, 0);
  ret = aclrtMemcpy(resultData2.data(), resultData2.size() * sizeof(resultData2[0]), outIndicesDeviceAddr,
                    size * sizeof(resultData2[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 < size2; i++) {
    LOG_PRINT("result indices [%ld] is: %ld\n", i, resultData2[i]);
  }

  // 6. 释放aclTensor和aclScalar，需要根据具体API的接口定义修改
  aclDestroyTensor(self);
  aclDestroyTensor(outValues);
  aclDestroyTensor(outIndices);
  return 0;
}

父主题： NN类算子接口