aclnnClampMaxTensor&aclnnInplaceClampMaxTensor
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
aclnnClampMaxTensor和aclnnInplaceClampMaxTensor实现相同的功能,使用区别如下,请根据自身实际场景选择合适的算子。
- aclnnClampMaxTensor:需新建一个输出张量对象存储计算结果。
- aclnnInplaceClampMaxTensor:无需新建输出张量对象,直接在输入张量的内存中存储计算结果。
每个算子分为两段式接口,必须先调用“aclnnClampMaxTensorGetWorkspaceSize”或者“aclnnInplaceClampMaxTensorGetWorkspaceSize”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“aclnnClampMaxTensor”或者“aclnnInplaceClampMaxTensor”接口执行计算。
aclnnStatus aclnnClampMaxTensorGetWorkspaceSize(const aclTensor* self, const aclTensor* max, aclTensor* out, uint64_t* workspaceSize, aclOpExecutor** executor)
aclnnStatus aclnnClampMaxTensor(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, aclrtStream stream)
aclnnStatus aclnnInplaceClampMaxTensorGetWorkspaceSize(aclTensor* selfRef, const aclTensor* max, uint64_t* workspaceSize, aclOpExecutor** executor)
aclnnStatus aclnnInplaceClampMaxTensor(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, aclrtStream stream)
功能描述
算子功能:将输入的所有元素限制在[-inf, max]范围内。
计算公式:
aclnnClampMaxTensorGetWorkspaceSize
参数说明:
- self(aclTensor*, 计算输入):输入tensor,数据类型支持FLOAT16、FLOAT、DOUBLE、INT8、UINT8、INT16、INT32、INT64、BFLOAT16(仅Atlas A2训练系列产品/Atlas 800I A2推理产品支持),且数据类型需要与max的数据类型需满足数据类型推导规则(参见互推导关系)。shape需要与max满足broadcast关系,支持非连续的Tensor,数据格式支持ND。
- max(aclTensor*, 计算输入):输入上限值tensor,数据类型支持FLOAT16、FLOAT、DOUBLE、INT8、UINT8、INT16、INT32、INT64、BFLOAT16(仅Atlas A2训练系列产品/Atlas 800I A2推理产品支持),且数据类型需要与self的数据类型需满足数据类型推导规则(参见互推导关系)。shape需要与self满足broadcast关系,支持非连续的Tensor,数据格式支持ND。
- out(aclTensor*, 计算输出):输出tensor,数据类型支持FLOAT16、FLOAT、DOUBLE、INT8、UINT8、INT16、INT32、INT64、BFLOAT16(仅Atlas A2训练系列产品/Atlas 800I A2推理产品支持),且数据类型需要是self与max推导之后可转换的数据类型,shape需要是self与max broadcast之后的shape,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。
161001 (ACLNN_ERR_PARAM_NULLPTR): 1. 传入的self、max或out为空指针。 161002 (ACLNN_ERR_PARAM_INVALID): 1. self与max推导后的数据类型不在支持范围内。 2. self与max的shape不满足broadcast关系,或broadcast后的shape与输出out的shape不一致。 3. self与max类型推导失败,或推导类型无法转为out的数据类型。 4. self、max或out的维度超过8维。
aclnnClampMaxTensor
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aclnnClampMaxTensorGetWorkspaceSize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。
aclnnInplaceClampMaxTensorGetWorkspaceSize
参数说明:
- selfRef(aclTensor*, 计算输入|计算输出):输入输出tensor,即公式中的self与out。数据类型支持FLOAT16、FLOAT、DOUBLE、INT8、UINT8、INT16、INT32、INT64、BFLOAT16(仅Atlas A2训练系列产品/Atlas 800I A2推理产品支持),且数据类型需要与max的数据类型需满足数据类型推导规则(参见互推导关系)。shape需要与max满足broadcast关系,支持非连续的Tensor,数据格式支持ND。
- max(aclTensor*, 计算输入):输入上限值tensor,数据类型支持FLOAT16、FLOAT、DOUBLE、INT8、UINT8、INT16、INT32、INT64、BFLOAT16(仅Atlas A2训练系列产品/Atlas 800I A2推理产品支持),且数据类型需要与selfRef的数据类型需满足数据类型推导规则(参见互推导关系)。shape需要与selfRef满足broadcast关系,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。
161001 (ACLNN_ERR_PARAM_NULLPTR): 1. 传入的selfRef或max为空指针。 161002 (ACLNN_ERR_PARAM_INVALID): 1. selfRef与max推导后的数据类型不在支持范围内。 2. selfRef与max的shape不满足broadcast 关系。 3. selfRef与max类型推导失败。 4. selfRef或max的维度超过8维。
aclnnInplaceClampMaxTensor
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aclnnInplaceClampMaxTensorGetWorkspaceSize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。
aclnnClampMaxTensor调用示例:
#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_clamp.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> maxShape = {4, 2};
std::vector<int64_t> outShape = {4, 2};
void* selfDeviceAddr = nullptr;
void* maxDeviceAddr = nullptr;
void* outDeviceAddr = nullptr;
aclTensor* self = nullptr;
aclTensor* max = nullptr;
aclTensor* out = nullptr;
std::vector<float> selfHostData = {0, 1, 2, 3, 4, 5, 6, 7};
std::vector<float> maxHostData = {1, 1, 1, 2, 2, 2, 3, 3};
std::vector<float> outHostData(8, 0);
// 创建self aclTensor
ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建max aclTensor
ret = CreateAclTensor(maxHostData, maxShape, &maxDeviceAddr, aclDataType::ACL_FLOAT, &max);
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aclnnClampMaxTensor第一段接口
ret = aclnnClampMaxTensorGetWorkspaceSize(self, max, out, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnClampMaxTensorGetWorkspaceSize 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aclnnClampMaxTensor第二段接口
ret = aclnnClampMaxTensor(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnClampMaxTensor 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]);
}
// 6. 释放aclTensor和aclScalar,需要根据具体API的接口定义修改
aclDestroyTensor(self);
aclDestroyTensor(max);
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(maxDeviceAddr);
aclrtFree(outDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}
aclnnInplaceClampMaxTensor调用示例:
#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_clamp.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> maxShape = {4, 2};
void* selfDeviceAddr = nullptr;
void* maxDeviceAddr = nullptr;
aclTensor* self = nullptr;
aclTensor* max = nullptr;
std::vector<float> selfHostData = {0, 1, 2, 3, 4, 5, 6, 7};
std::vector<float> maxHostData = {1, 1, 1, 2, 2, 2, 3, 3};
// 创建self aclTensor
ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建max aclTensor
ret = CreateAclTensor(maxHostData, maxShape, &maxDeviceAddr, aclDataType::ACL_FLOAT, &max);
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aclnnInplaceClampMaxTensor第一段接口
ret = aclnnInplaceClampMaxTensorGetWorkspaceSize(self, max, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnInplaceClampMaxTensorGetWorkspaceSize 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aclnnInplaceClampMaxTensor第二段接口
ret = aclnnInplaceClampMaxTensor(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnInplaceClampMaxTensor 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(selfShape);
std::vector<float> resultData(size, 0);
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(max);
// 7. 释放device资源,需要根据具体API的接口定义修改
aclrtFree(selfDeviceAddr);
aclrtFree(maxDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}