下载
中文
注册
我要评分
文档获取效率
文档正确性
内容完整性
文档易理解
在线提单
论坛求助
昇腾小AI

aclnnScale

支持的产品型号

  • Atlas 推理系列产品。
  • Atlas A2训练系列产品/Atlas 800I A2推理产品。

接口原型

每个算子分为两段式接口,必须先调用“aclnnScaleGetWorkspaceSize”接口获取计算所需workspace大小以及包含了算子计算流程的执行器,再调用“aclnnScale”接口执行计算。

  • aclnnstatus aclnnScaleGetWorkspaceSize(const aclTensor *x, const aclTensor *scale, const aclTensor *bias, int64_t axis, int64_t numAxes, bool scaleFromBlob, aclTensor *y, uint64_t *workspaceSize, aclOpExecutor **executor)
  • aclnnstatus aclnnScale(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream)

功能描述

计算公式:

若不输入bias,则

y=xscaley=x*scale

若输入bias,则

y=xscale+biasy=x*scale + bias

说明: scale/bias支持跟X的broadcast, scale/bias shape规则 scaleFromBlob is True(axis 转换为正数, numAxes = -1 表示到最后轴) scaleShape = xShape[axis:axis + numAxes] biasShape = xShape[axis:axis + numAxes]

scaleFromBlob is False(axis 转换为正数, numAxes = -1 表示到最前轴) scaleShape must be xShape[axis:axis + rank(scaleShape)] biasShape must be xShape[axis:axis + rank(scaleShape)] ex: scaleFromBlob = True xShape = [a, b, c, d, e, f] axis = 3 numAxes = 2 --> scaleShape = [d, e] xShape = [a, b, c, d, e, f] axis = 3 numAxes = 3 --> scaleShape = [d, e, f] xShape = [a, b, c, d, e, f] axis = 3 numAxes = -1 --> scaleShape = [d, e, f] scaleFromBlob = False xShape = [a, b, c, d, e, f] axis = 3 --> scaleShape = [d, e] xShape = [a, b, c, d, e, f] axis = 3 --> scaleShape = [d, e, f] xShape = [a, b, c, d, e, f] axis = 3 --> scaleShape = [d] xShape = [a, b, c, d, e, f] axis = 3 --> scaleShape = [e] [错误]

aclnnScaleGetWorkspaceSize

  • 参数说明:

    • x(aclTensor*, 计算输入): 算子输入的Tensor,Device侧的aclTensor,数据类型支持FLOAT32, FLOAT16、BFLOAT16(Atlas 推理系列产品不支持),支持非连续的Tensor数据格式支持ND.
    • scale(aclTensor*, 计算输入): 算子输入的Tensor,Device侧的aclTensor,数据类型支持FLOAT32, FLOAT16、BFLOAT16(Atlas 推理系列产品不支持),数据类型需要与x的数据类型相同,支持非连续的Tensor数据格式支持ND.shape满足broadcast要求,见功能描述章节说明.
    • bias(aclTensor*, 计算输入): 算子输入的Tensor,Device侧的aclTensor,数据类型支持FLOAT32, FLOAT16、BFLOAT16(Atlas 推理系列产品不支持),不为空时数据类型需要与scale的数据类型相同,支持非连续的Tensor数据格式支持ND.shape与scale保持一致.
    • axis(int64_t, 计算输入) host侧INT64类型,指定进行scale的起始轴, 取值范围 [-x_rank, x_rank)(x_rank表示x的shape维度).
    • numAxes(int64_t, 计算输入) host侧INT64类型,指定进行scale的轴长度, 取值范围 >= -1, numAxes = -1, 表示从axis轴开始scale到最后一轴.
    • scaleFromBlob(bool, 计算输入) host侧BOOL类型,指定要scaleFromBlob类型, True: scale from blob, 使用numAxes + axis进行scale, False: scale from input scale, 从axis开始 scale input scale长度, 忽略numAxes取值.
    • y(aclTensor*, 计算输出): 输出Tensor,shape维度和x保持一致,Device侧的aclTensor,数据类型支持FLOAT32, FLOAT16、BFLOAT16(Atlas 推理系列产品不支持),数据类型需要与x的数据类型相同,支持非连续的Tensor数据格式支持ND.
    • workspaceSize(uint64_t*, 出参): 返回需要在Device侧申请的workspace大小.
    • executor(aclOpExecutor**, 出参): 返回op执行器,包含了算子计算流程.
  • 返回值:

    aclnnStatus: 返回状态码,具体参见aclnn返回码.

    返回161001(ACLNN_ERR_PARAM_NULLPTR):1. 传入的x、scale、y是空指针。
    返回161002(ACLNN_ERR_PARAM_INVALID):1. x的数据类型不在支持的范围之内。
                                          2. bias不为空时,bias与scale的数据类型不一致。
                                          3. scale与x的数据类型不一致。
                                          4. y与x的数据类型不一致。
                                          5. x和y的shape不一致。
                                          6. bias不为空时,bias与scale的shape不一致。
                                          7. x和scale的shape维度大于8.
                                          8. axis的取值不在[-x_rank, x_rank)范围内。
                                          9. numAxes的取值小于-1。
                                          10. scaleFromBlob为True,numAxes等于0且scale的shape不为[1]。
                                          11. axis转换为正数之后与numAxes相加,大于x_rank。
                                          12. scale的shape与预期不符(预期shape推导参考功能描述)。

aclnnScale

  • 参数说明:

    • workspace(void *, 入参):在Device侧申请的workspace内存地址。
    • workspaceSize(uint64_t, 入参):在Device侧申请的workspace大小,由第一段接口aclnnScaleGetWorkspaceSize获取。
    • executor(aclOpExecutor *, 入参):op执行器,包含了算子计算流程。
    • stream(aclrtStream, 入参):指定执行任务的 AscendCL Stream流。
  • 返回值:

    aclnnStatus:返回状态码,具体参见aclnn返回码

约束与限制

Atlas 推理系列产品不支持scale、offset及输入为bf16。

调用示例

示例代码如下,仅供参考,具体编译和执行过程请参考编译与运行样例

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/level2/aclnn_scale.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> tensor1Shape = {4};
  std::vector<int64_t> tensor2Shape = {4};
  std::vector<int64_t> outShape = {4, 2};
  void* selfDeviceAddr = nullptr;
  void* tensor1DeviceAddr = nullptr;
  void* tensor2DeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* self = nullptr;
  aclTensor* tensor1 = nullptr;
  aclTensor* tensor2 = nullptr;
  aclTensor* out = nullptr;

  std::vector<float> selfHostData = {0, 1, 2, 3, 4, 5, 6, 7};
  std::vector<float> tensor1HostData = {2, 2, 2, 2, 2, 2, 2, 2};
  std::vector<float> tensor2HostData = {2, 2, 2, 2, 2, 2, 2, 2};
  std::vector<float> outHostData = {0, 0, 0, 0, 0, 0, 0, 0};
  int64_t axis = 0;
  int64_t numAxes = 1;
  bool fromBlom = true;

  // 创建self aclTensor
  ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 创建tensor1 aclTensor
  ret = CreateAclTensor(tensor1HostData, tensor1Shape, &tensor1DeviceAddr, aclDataType::ACL_FLOAT, &tensor1);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 创建tensor2 aclTensor
  ret = CreateAclTensor(tensor2HostData, tensor2Shape, &tensor2DeviceAddr, aclDataType::ACL_FLOAT, &tensor2);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 创建out aclTensor
  ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
  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aclnnScale第一段接口
  ret = aclnnScaleGetWorkspaceSize(self, tensor1, tensor2, axis, numAxes, fromBlom,  out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnScaleGetWorkspaceSize 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aclnnScale第二段接口
  ret = aclnnScale(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnScale 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 resultData from device to host failed. ERROR: %d\n", ret);
            return ret);
  for (int64_t i = 0; i < size; i++) {
    LOG_PRINT("resultData[%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(tensor1);
  aclDestroyTensor(tensor2);
  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(tensor1DeviceAddr);
  aclrtFree(tensor2DeviceAddr);
  aclrtFree(outDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}
搜索结果
找到“0”个结果

当前产品无相关内容

未找到相关内容,请尝试其他搜索词