aclnnStdMeanCorrection
支持的产品型号
- Atlas 训练系列产品。
- 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“aclnnStdMeanCorrectionGetWorkspaceSize”接口获取入参并根据计算流程计算所需workspace大小,再调用“aclnnStdMeanCorrection”接口执行计算。
aclnnStatus aclnnStdMeanCorrectionGetWorkspaceSize(const aclTensor* self, const aclIntArray* dim, int64_t correction, bool keepdim, aclTensor* stdOut, aclTensor* meanOut, uint64_t* workspaceSize, aclOpExecutor** executor)
aclnnStatus aclnnStdMeanCorrection(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, const aclrtStream stream)
功能描述
- 算子功能:计算样本标准差和均值。
- 计算公式:
假设 dim 为 ,则对该维度进行计算。为该维度的 shape。取 ,求出该维度上的平均值 。
当
keepdim = true
时,reduce后保留该维度,且输出shape中该维度值为1;当keepdim = false
时,不保留。
aclnnStdMeanCorrectionGetWorkspaceSize
参数说明
- self(aclTensor*, 计算输入):公式中的
self
,Device侧的aclTensor,支持非连续的Tensor。 - dim(aclIntArray*, 计算输入):公式中的
dim
,Host侧的aclIntArray,参与计算的维度,取值范围为[-self.dim(), self.dim()-1],支持的数据类型为INT32、INT64。 - correction(int64_t*, 计算输入):公式中
$\delta N$
,Host侧的整型,修正值。 - keepdim(bool*, 计算输入):公式中
keepdim
,Host侧的布尔型,是否在输出张量中保留输入张量的维度。 - stdOut(aclTensor*, 计算输出):公式中
stdOut
,Device侧的aclTensor,支持非连续的Tensor。 - meanOut(aclTensor*, 计算输出):公式中
meanOut
,Device侧的aclTensor,支持非连续的Tensor。 - 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。
- self(aclTensor*, 计算输入):公式中的
返回值:
aclnnStatus: 返回状态码,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_INVALID):1. 传入的 self、stdOut、meanOut是空指针时。 返回161002(ACLNN_ERR_PARAM_INVALID):1. self、stdOut、meanOut数据类型不在支持的范围之内。 2. dim 数组中的维度超出 self 的维度范围。 3. dim 数组中元素重复。 4. stdOut的shape出现如下情况会出错: keepdim为true时,stdOut.shape != self.shape(指定维度dim设置为1的形状); keepdim为false时,stdOut.shape != self.shape(去除指定维度dim后的形状) 5. meanOut的shape出现如下情况会出错: keepdim为true时,meanOut.shape != self.shape(指定维度dim设置为1的形状); keepdim为false时,meanOut.shape != self.shape(去除指定维度dim后的形状)
aclnnStdMeanCorrection
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aclnnStdMeanCorrectionGetWorkspaceSize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
无
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/aclnn_std_mean_correction.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, 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. 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 = {2, 3, 4};
std::vector<int64_t> stdOutShape = {2, 4};
std::vector<int64_t> meanOutShape = {2, 4};
void* selfDeviceAddr = nullptr;
void* stdOutDeviceAddr = nullptr;
void* meanOutDeviceAddr = nullptr;
aclTensor* self = nullptr;
aclTensor* stdOut = nullptr;
aclTensor* meanOut = nullptr;
std::vector<float> selfHostData = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18,
19, 20, 21, 22, 23, 24};
std::vector<float> stdOutHostData = {1, 2, 3, 4, 5, 6, 7, 8.0};
std::vector<float> meanOutHostData = {1, 2, 3, 4, 5, 6, 7, 8.0};
std::vector<int64_t> dimData = {1};
int64_t correction = 1;
bool keepdim = false;
// 创建self aclTensor
ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建stdOut aclTensor
ret = CreateAclTensor(stdOutHostData, stdOutShape, &stdOutDeviceAddr, aclDataType::ACL_FLOAT, &stdOut);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建meanOut aclTensor
ret = CreateAclTensor(meanOutHostData, meanOutShape, &meanOutDeviceAddr, aclDataType::ACL_FLOAT, &meanOut);
CHECK_RET(ret == ACL_SUCCESS, return ret);
const aclIntArray *dim = aclCreateIntArray(dimData.data(), dimData.size());
CHECK_RET(dim != nullptr, return ACL_ERROR_INTERNAL_ERROR);
// 3. 调用CANN算子库API,需要修改为具体的API
uint64_t workspaceSize = 0;
aclOpExecutor* executor;
// 调用aclnnStdMeanCorrection第一段接口
ret = aclnnStdMeanCorrectionGetWorkspaceSize(self, dim, correction, keepdim, stdOut, meanOut, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnStdMeanCorrectionGetWorkspaceSize 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aclnnStdMeanCorrection第二段接口
ret = aclnnStdMeanCorrection(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnStdMeanCorrection 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(stdOutShape);
std::vector<float> stdResultData(size, 0);
ret = aclrtMemcpy(stdResultData.data(), stdResultData.size() * sizeof(stdResultData[0]), stdOutDeviceAddr, size * sizeof(float),
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("stdResultData[%ld] is: %f\n", i, stdResultData[i]);
}
std::vector<float> meanResultData(size, 0);
ret = aclrtMemcpy(meanResultData.data(), meanResultData.size() * sizeof(meanResultData[0]), meanOutDeviceAddr, size * sizeof(float),
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("meanResultData[%ld] is: %f\n", i, meanResultData[i]);
}
// 6. 释放aclTensor和aclScalar,需要根据具体API的接口定义修改
aclDestroyTensor(self);
aclDestroyTensor(stdOut);
aclDestroyTensor(meanOut);
// 7. 释放device资源,需要根据具体API的接口定义修改
aclrtFree(selfDeviceAddr);
aclrtFree(stdOutDeviceAddr);
aclrtFree(meanOutDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}