aclnnBincount

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 aclnnBincountGetWorkspaceSize(const aclTensor *self, const aclTensor * weights, int64_t minlength, aclTensor *out, uint64_t *workspaceSize, aclOpExecutor **executor)
• 第二段接口：aclnnStatus aclnnBincount(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, const aclrtStream stream)

功能描述

• 算子功能：计算非负整数数组中每个值的频率。

  当weights为空时，self中self[i]对应的数每出现1次，则其频率加1；当weights不为空时，self中self[i]对应的数每出现1次，则其频率加weights[i]，最后存放到out的第self[i]+1 位置上；因此out大小为self 中最大值+1。

• 计算公式：

  如果self_i是self位置i上的值，如果指定了weights，公式如下：

  

  否则公式如下：

  

aclnnBincountGetWorkspaceSize

• 接口定义：

  aclnnStatus aclnnBincountGetWorkspaceSize(const aclTensor *self, const aclTensor * weights, int64_t minlength, aclTensor *out, uint64_t *workspaceSize, aclOpExecutor **executor)

• 参数说明：
  • self：npu device侧aclTensor，输入张量，数据类型支持INT8、INT16、INT32、INT64、UINT8，且必须是非负整数。支持非连续的Tensor，数据格式支持1维ND。
  • weights：npu device侧aclTensor，self每个值的权重，可为空指针。数据类型支持FLOAT16、FLOAT、DOUBLE、INT8、INT16、INT64、INT32、UINT8、BOOL。支持非连续的Tensor，数据格式支持1维ND，且shape需要与self一致。
  • minlength：host侧int类型，指定输出最小长度。如果计算出来的self最大值小于minlength，则输出长度为minlength。数据类型支持INT64型。
  • out：npu device侧aclTensor，输出张量，数据类型支持FLOAT、INT32、INT64、DOUBLE，数据格式为1维ND，支持非连续的Tensor。
  • workspaceSize：返回用户需要在npu 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是空指针。
  • 返回161002（ACLNN_ERR_PARAM_INVALID）：
    • self、weights的数据类型和数据格式不在支持的范围内。
    • 当weights不为空时，self、weights的shape不一致。

aclnnBincount

• 接口定义：

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

• 参数说明：
  • workspace：在npu device侧申请的workspace内存起址。
  • workspaceSize：在npu device侧申请的workspace大小，由第一段接口aclnnBincountGetWorkspaceSize获取。
  • executor：op执行器，包含了算子计算流程。
  • stream：acl stream流。
• 返回值：

  返回aclnnStatus状态码，具体参见aclnn返回码。

约束与限制

无

调用示例

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/level2/aclnn_max.h"
#include "aclnnop/level2/aclnn_bincount.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 shape_size = 1;
  for (auto i : shape) {
    shape_size *= i;
  }
  return shape_size;
}

int Init(int32_t deviceId, aclrtContext* context, aclrtStream* stream) {
  // 固定写法，acl初始化
  auto 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);
  ret = aclInit(nullptr);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit 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acl对外接口列表
  // 根据自己的实际device填写deviceId
  int32_t deviceId = 0;
  aclrtContext contextMax;
  aclrtStream streamMax;
  auto ret = Init(deviceId, &contextMax, &streamMax);
  // check根据自己的需要处理
  CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);

  // 2. 先调max获取self中的最大值，然后使用该最大值与minlength比较最大值去获得输出tensor的size
  // 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 = {8};
  std::vector<int64_t> maxOutShape={1};
  std::vector<int32_t> selfHostData = {8,1,2,3,4,5,6,7};
  std::vector<int32_t> maxOutHostData(1, 0);
  void* selfDeviceAddr = nullptr;
  void* maxOutDeviceAddr = nullptr;
  aclTensor* self = nullptr;
  aclTensor* maxOut = nullptr;
  ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_INT32, &self);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(maxOutHostData, maxOutShape, &maxOutDeviceAddr, aclDataType::ACL_INT32, &maxOut);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 调用CANN算子库API，需要修改为具体的HostApi
  uint64_t workspaceSizeMax = 0;
  aclOpExecutor* executorMax;
  // 调用aclnnMax第一段接口
  ret = aclnnMaxGetWorkspaceSize(self, maxOut, &workspaceSizeMax, &executorMax);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMaxGetWorkspaceSize 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* workspaceAddrMax = nullptr;
  if (workspaceSizeMax > 0) {
    ret = aclrtMalloc(&workspaceAddrMax, workspaceSizeMax, ACL_MEM_MALLOC_HUGE_FIRST);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret;);
  }
  // 调用aclnnMax第二段接口
  ret = aclnnMax(workspaceAddrMax, workspaceSizeMax, executorMax, streamMax);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMax failed. ERROR: %d\n", ret); return ret);
  // （固定写法）同步等待任务执行结束
  ret = aclrtSynchronizeStream(streamMax);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
  // 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
  std::vector<int32_t> resultDataMax(1, 0);
  ret = aclrtMemcpy(resultDataMax.data(), sizeof(resultDataMax[0]), maxOutDeviceAddr, sizeof(int32_t), 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);
  // 释放aclTensor和aclScalar，需要根据具体API的接口定义修改
  aclDestroyTensor(maxOut);

  // 3. 构造输入与输出，需要根据API的接口自定义构造
  int64_t minlength = 0;
  int64_t outSize = (resultDataMax[0] < minlength) ? minlength : resultDataMax[0] + 1;
  std::vector<int64_t> weightsShape = {8};
  std::vector<int64_t> outShape = {outSize};
  void* weightsDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* weights = nullptr;
  aclTensor* out = nullptr;
  std::vector<float> weightsHostData = {1, 1, 1.1, 2, 2, 2, 3, 3};
  std::vector<float> outHostData(outSize, 0);
  ret = CreateAclTensor(weightsHostData, weightsShape, &weightsDeviceAddr, aclDataType::ACL_FLOAT, &weights);
  CHECK_RET(ret == ACL_SUCCESS, return ret); 
  ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 调用CANN算子库API，需要修改为具体的HostApi
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // 调用aclnnBincount第一段接口
  ret = aclnnBincountGetWorkspaceSize(self, weights, minlength, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnBincountGetWorkspaceSize 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aclnnBincount第二段接口
  ret = aclnnBincount(workspaceAddr, workspaceSize, executor, streamMax);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnBincount failed. ERROR: %d\n", ret); return ret);

  // 4. （固定写法）同步等待任务执行结束
  ret = aclrtSynchronizeStream(streamMax);
  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, 55);
  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. 释放aclTensor和aclScalar，需要根据具体API的接口定义修改
  aclDestroyTensor(self);
  aclDestroyTensor(weights);
  aclDestroyTensor(out);
  return 0;
}