1 AOL算子库

CANN（Compute Architecture for Neural Networks）提供了算子加速库（Ascend Operator Library，简称AOL）。该库提供了一系列丰富且深度优化过的高性能算子API，更亲和昇腾AI处理器，调用流程如图1所示。开发者可直接调用算子库API使能模型创新与应用，以进一步提升开发效率和获取极致模型性能。

单算子API执行的算子接口一般定义为“两段式接口”，以NN算子接口定义为例：

aclnnStatus aclnnXxxGetWorkspaceSize(const aclTensor *src, ..., aclTensor *out, ..., uint64_t *workspaceSize, aclOpExecutor **executor);
aclnnStatus aclnnXxx(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream);

其中aclnnXxxGetWorkspaceSize为第一段接口，主要用于计算本次API调用计算过程中需要多少的workspace内存。获取到本次API计算需要的workspace大小后，按照workspaceSize大小申请AI处理器内存，然后调用第二段接口aclnnXxx。
说明：

• workspace是指除输入/输出外，API在AI处理器上完成计算所需要的临时内存。
• 第二段接口aclnnXxx(…)不能重复调用，如下调用方式会出现异常：
  aclnnXxxGetWorkspaceSize(…)
  aclnnXxx(…)
  aclnnXxx(…)

2 具体示例

2.1 文件准备

可以从官网获得一个算子的使用示例，如下算子是aclnnAdd：

aclnnAdd&aclnnInplaceAdd-单算子接口-NN类算子接口-单算子API执行-单算子执行-AscendCL API（C&C++）-应用开发接口-CANN商用版8.0.RC2.2开发文档-昇腾社区
​

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_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) {
  // 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 = 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/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;
  aclrtStream stream;
  auto ret = Init(deviceId, &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
  std::vector<int64_t> selfShape = {4, 2};
  std::vector<int64_t> otherShape = {4, 2};
  std::vector<int64_t> outShape = {4, 2};
  void* selfDeviceAddr = nullptr;
  void* otherDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* self = nullptr;
  aclTensor* other = nullptr;
  aclScalar* alpha = nullptr;
  aclTensor* out = nullptr;
  std::vector<float> selfHostData = {0, 1, 2, 3, 4, 5, 6, 7};
  std::vector<float> otherHostData = {1, 1, 1, 2, 2, 2, 3, 3};
  std::vector<float> outHostData(8, 0);
  float alphaValue = 1.2f;
  // 创建self aclTensor
  ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 创建other aclTensor
  ret = CreateAclTensor(otherHostData, otherShape, &otherDeviceAddr, aclDataType::ACL_FLOAT, &other);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 创建alpha aclScalar
  alpha = aclCreateScalar(&alphaValue, aclDataType::ACL_FLOAT);
  CHECK_RET(alpha != nullptr, return ret);
  // 创建out aclTensor
  ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  
  // aclnnAdd接口调用示例  
  // 3. 调用CANN算子库API
  // 调用aclnnAdd第一段接口
  ret = aclnnAddGetWorkspaceSize(self, other, alpha, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnAddGetWorkspaceSize 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aclnnAdd第二段接口
  ret = aclnnAdd(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnAdd 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(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]);
  }

    
  // aclnnInplaceAdd接口调用示例  
  // 3. 调用CANN算子库API
  LOG_PRINT("\ntest aclnnInplaceAdd\n");
  // 调用aclnnInplaceAdd第一段接口
  ret = aclnnInplaceAddGetWorkspaceSize(self, other, alpha, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnInplaceAddGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
  // 根据第一段接口计算出的workspaceSize申请device内存
  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aclnnInplaceAdd第二段接口
  ret = aclnnInplaceAdd(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnInplaceAdd 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
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), selfDeviceAddr,
                    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. 释放aclTensor和aclScalar，需要根据具体API的接口定义修改
  aclDestroyTensor(self);
  aclDestroyTensor(other);
  aclDestroyScalar(alpha);
  aclDestroyTensor(out);

  // 7. 释放Device资源，需要根据具体API的接口定义修改
  aclrtFree(selfDeviceAddr);
  aclrtFree(otherDeviceAddr);
  aclrtFree(outDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}

如果将该文件命名为test_add.cpp，那么接下来写它的CMakeLists文件，可从如下模板中修改。

重要内容修改：

• add_executable中的文件名称，比如当前要改成test_add.cpp

# Copyright (c) Huawei Technologies Co., Ltd. 2019. All rights reserved.

# CMake lowest version requirement
cmake_minimum_required(VERSION 3.14)

# 设置工程名
project(ACLNN_EXAMPLE)

# Compile options
add_compile_options(-std=c++11)

# 设置编译选项
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY  "./bin")    
set(CMAKE_CXX_FLAGS_DEBUG "-fPIC -O0 -g -Wall")
set(CMAKE_CXX_FLAGS_RELEASE "-fPIC -O2 -Wall")

# 设置可执行文件名（如opapi_test），并指定待运行算子文件*.cpp所在目录
add_executable(opapi_test
               test_add.cpp) 

# 设置ASCEND_PATH（CANN软件包目录，请根据实际路径修改）和INCLUDE_BASE_DIR（头文件目录）
if(NOT "$ENV{ASCEND_CUSTOM_PATH}" STREQUAL "")      
    set(ASCEND_PATH $ENV{ASCEND_CUSTOM_PATH})
else()
    set(ASCEND_PATH "/usr/local/Ascend/ascend-toolkit/latest")
endif()
set(INCLUDE_BASE_DIR "${ASCEND_PATH}/include")
include_directories(
    ${INCLUDE_BASE_DIR}
    ${INCLUDE_BASE_DIR}/aclnn
)

# 设置链接的库文件路径
target_link_libraries(opapi_test PRIVATE
                      ${ASCEND_PATH}/lib64/libascendcl.so
                      ${ASCEND_PATH}/lib64/libnnopbase.so
                      ${ASCEND_PATH}/lib64/libopapi.so)


# 可执行文件在CMakeLists文件所在目录的bin目录下
install(TARGETS opapi_test DESTINATION ${CMAKE_RUNTIME_OUTPUT_DIRECTORY})

2.2 编译运行

1、进入CMakeLists.txt所在目录，执行如下命令，新建build目录存放生成的编译文件。执行：

source ${install_path}/set_env.sh。#install_path为CANN的安装目录，一般为/usr/local/Ascend/ascend-toolkit/latest

2、进入build目录，执行cmake命令编译，再执行make命令生成可执行文件。

cd build
cmake ../ -DCMAKE_CXX_COMPILER=g++ -DCMAKE_SKIP_RPATH=TRUE
make

编译成功后，会在build目录的bin文件夹下生成opapi_test可执行文件。

3、进入bin目录，运行可执行文件opapi_test。

cd bin
./opapi_test

以Add算子的运行结果为例，运行后的结果示例如下：

result[0] is: 1.200000
result[1] is: 2.200000
result[2] is: 3.200000
result[3] is: 5.400000
result[4] is: 6.400000
result[5] is: 7.400000
result[6] is: 9.600000
result[7] is: 10.600000

可参考官网：

编译与运行样例-NN类算子接口-单算子API执行-单算子执行-AscendCL API（C&C++）-应用开发接口-API参考-CANN商用版8.0.RC2.2开发文档-昇腾社区 (hiascend.com)