aclnnAdaptiveAvgPool3dBackward
支持的产品型号
- 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“aclnnAdaptiveAvgPool3dBackwardGetWorkspaceSize”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“aclnnAdaptiveAvgPool3dBackward”接口执行计算。
aclnnStatus aclnnAdaptiveAvgPool3dBackwardGetWorkspaceSize(const aclTensor* gradOutput, const aclTensor* self, aclTensor* out, uint64_t* workspaceSize, aclOpExecutor** executor)
aclnnStatus aclnnAdaptiveAvgPool3dBackward(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, aclrtStream stream)
功能描述
- 算子功能:进行aclnnAdaptiveAvgPool3d api的结果的反向计算。
aclnnAdaptiveAvgPool3dBackwardGetWorkspaceSize
参数说明:
- gradOutput(aclTensor*,计算输入):当前节点的梯度,Device侧的aclTensor。数据类型支持BFLOAT16、FLOAT16、FLOAT32,且数据类型与self一致。支持非连续的Tensor,shape支持4维或5维,shape的每一维均为正数,且总维数与self一致。数据格式支持NCDHW、ND,且需要与self数据格式一致。
- self(aclTensor*, 计算输入):输入张量,叶子节点。Device侧的aclTensor,数据类型支持BFLOAT16、FLOAT16、FLOAT32,支持非连续的Tensor,shape支持4维或5维,且shape的每一维均为正数。数据格式支持NCDHW、ND。
- out(aclTensor*, 计算输出):输出张量,对应了输入叶子节点的梯度。Device侧的aclTensor,数据类型支持BFLOAT16、FLOAT16、FLOAT32;shape与self保持一致;数据格式支持NCDHW、ND,且与self数据类型一致。
- workspaceSize(uint64_t*,出参):返回用户需要在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。
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):1. 传入的gradOutput、self或out是空指针。 返回161002 (ACLNN_ERR_PARAM_INVALID):1. gradOutput和self的数据类型和数据格式不在支持的范围之内。 2. gradOutput、self和out数据类型不一致。 3. gradOutput、self和out的维数不等于4或5。 4. gradOutput、self和out的shape不匹配。 5. gradOutput或self的shape的某一维不大于0。 6. gradOutput和self的数据格式不一致。
aclnnAdaptiveAvgPool3dBackward
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aclnnAdaptiveAvgPool3dBackwardGetWorkspaceSize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 "math.h"
#include "acl/acl.h"
#include "aclnnop/aclnn_adaptive_avg_pool3d_backward.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. device/stream初始化,参考AscendCL对外接口列表
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
std::vector<int64_t> yGradShape = {2, 2, 1, 1, 2};
std::vector<int64_t> xShape = {2, 2, 1, 1, 4};
std::vector<int64_t> xGradShape = {2, 2, 1, 1, 4};
void* yGradDeviceAddr = nullptr;
void* xDeviceAddr = nullptr;
void* xGradDeviceAddr = nullptr;
aclTensor* yGrad = nullptr;
aclTensor* x = nullptr;
aclTensor* xGrad = nullptr;
std::vector<float> yGradHostData = {1, 2, 3, 4, 5, 6, 7, 8};
std::vector<float> xHostData(GetShapeSize(xShape), 1);
std::vector<float> xGradHostData(16, 0);
// 创建yGrad aclTensor
ret = CreateAclTensor(yGradHostData, yGradShape, &yGradDeviceAddr, aclDataType::ACL_FLOAT, &yGrad);
// 创建x aclTensor
ret = CreateAclTensor(xHostData, xShape, &xDeviceAddr, aclDataType::ACL_FLOAT, &x);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建xGrad aclTensor
ret = CreateAclTensor(xGradHostData, xGradShape, &xGradDeviceAddr, aclDataType::ACL_FLOAT, &xGrad);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 3. 调用CANN算子库API
uint64_t workspaceSize = 0;
aclOpExecutor* executor;
// 调用aclnnAdaptiveAvgPool3dBackward第一段接口
ret = aclnnAdaptiveAvgPool3dBackwardGetWorkspaceSize(yGrad, x, xGrad, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnAdaptiveAvgPool3dBackwardGetWorkspaceSize 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aclnnAdaptiveAvgPool3dBackward二段接口
ret = aclnnAdaptiveAvgPool3dBackward(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnAdaptiveAvgPool3dBackward failed. ERROR: %d\n", ret); return ret);
// 4. 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
auto size = GetShapeSize(xGradShape);
std::vector<float> resultData(size, 0);
ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), xGradDeviceAddr,
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
aclDestroyTensor(yGrad);
aclDestroyTensor(x);
aclDestroyTensor(xGrad);
// 7. 释放Device资源,需要根据具体API的接口定义修改
aclrtFree(yGradDeviceAddr);
aclrtFree(xDeviceAddr);
aclrtFree(xGradDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}