CANN/ops-nn PReLU反向传播

aclnnPreluBackward【免费下载链接】ops-nn本项目是CANN提供的神经网络类计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-nn 查看源码产品支持情况产品是否支持Ascend 950PR/Ascend 950DT√Atlas A3 训练系列产品/Atlas A3 推理系列产品√Atlas A2 训练系列产品/Atlas A2 推理系列产品√Atlas 200I/500 A2 推理产品×Atlas 推理系列产品×Atlas 训练系列产品√功能说明接口功能完成aclnnPreluBackward的反向计算。gradInput的计算公式如下$$ gradInput_{i,j,...} \begin{cases} gradOutput_{i,j,...}, if\ self_{i,j,...} 0 \ gradOutput_{i,j,...} * weight_{i}, if\ self_{i,j,...} 0 \end{cases} $$gradWeight的计算公式如下$$ gradWeight_{j}\sum_{i,...} \begin{cases} 0, if\ self_{i,j,...} 0 \ gradOutput_{i,j,...} * self_{i,j,...}, if\ self_{i,j,...} 0 \end{cases} $$函数原型每个算子分为两段式接口必须先调用“aclnnPreluBackwardGetWorkspaceSize”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“aclnnPreluBackward”接口执行计算。aclnnStatus aclnnPreluBackwardGetWorkspaceSize( const aclTensor* gradOutput, const aclTensor* self, const aclTensor* weight, aclTensor* gradInput, aclTensor* gradWeight, uint64_t* workspaceSize, aclOpExecutor** executor)aclnnStatus aclnnPreluBackward( void* workspace, uint64_t workspace_size, aclOpExecutor* executor, aclrtStream stream)aclnnPreluBackwardGetWorkspaceSize参数说明参数名输入/输出描述使用说明数据类型数据格式维度(shape)非连续TensorgradOutputaclTensor*输入反向传播的梯度值。公式中的gradOutput。支持空Tensor。dtype需要与self相同。shape需要与self满足broadcast关系且Broadcast后shape与self的shape相等。FLOAT16、FLOAT32、BFLOAT16ND0-8√selfaclTensor*输入prelu的正向输入值。公式中的self。支持空Tensor。FLOAT16、FLOAT32、BFLOAT16ND0-8√weightaclTensor*输入prelu的权重公式中的weight。支持空Tensor。dtype需要与self相同。当self的shape维度大于1维时weight的shape维度可以与self的shape维度相同且第2维度的值保持一致同时weight的shape其他维度的值为1或者weight是1维Tensor元素个数为self的shape的第2维度。否则weight元素个数为1。FLOAT16、FLOAT32、BFLOAT16ND0-8√gradInputaclTensor*输出为self的梯度值。dtype需要与self相同。shape需要与gradOutput满足broadcast关系。gradInput的shape和数据类型与self的相同。FLOAT、FLOAT16、BFLOAT16ND0-8√gradWeightaclTensor*输出为weight的梯度值。dtype需要与self相同。需要与weight的数据类型相同。gradWeight的shape与weight的shape保持一致。FLOAT、FLOAT16、BFLOAT16ND0-8√workspaceSizeuint64_t*输出返回需要在Device侧申请的workspace大小。-----executoraclOpExecutor**输出返回op执行器包含了算子计算流程。-----Atlas 训练系列产品 数据类型支持FLOAT16、FLOAT32。返回值aclnnStatus返回状态码具体参见aclnn返回码。第一段接口会完成入参校验出现以下场景时报错返回码错误码描述ACLNN_ERR_PARAM_NULLPTR161001传入的gradOutput、self、weight、gradInput、gradWeight是空指针。ACLNN_ERR_PARAM_INVALID161002gradOutput、self、weight、gradInput、gradWeight的数据类型不在支持的范围之内。gradOutput、self、weight、gradInput、gradWeight的数据类型不同。gradOutput、self、weight、gradInput、gradWeight大于8维。weight的元素个数不等于self的通道数或者1。weight的元素个数为1时gradWeight的shape与weight不相同。gradOutput和self的shape不满足条件broadcastshape条件。aclnnPreluBackward参数说明参数名输入/输出描述workspace输入在Device侧申请的workspace内存地址。workspaceSize输入在Device侧申请的workspace大小由第一段接口aclnnPreluBackwardGetWorkspaceSize获取。executor输入op执行器包含了算子计算流程。stream输入指定执行任务的Stream。返回值aclnnStatus返回状态码具体参见aclnn返回码。约束说明确定性计算aclnnPreluBackward默认确定性实现。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_prelu_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::vectorint64_t shape) { int64_t shape_size 1; for (auto i : shape) { shape_size * i; } return shape_size; } int Init(int32_t deviceId, aclrtStream* stream) { // 固定写法资源初始化 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::vectorT hostData, const std::vectorint64_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::vectorint64_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acl API手册 // 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 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::vectorint64_t selfShape {4, 2}; std::vectorint64_t weightShape {2}; std::vectorint64_t gradOutputShape {4, 2}; std::vectorint64_t gradInputShape {4, 2}; std::vectorint64_t gradWeightShape {2}; void* selfDeviceAddr nullptr; void* gradOutputDeviceAddr nullptr; void* weightDeviceAddr nullptr; void* gradInputDeviceAddr nullptr; void* gradWeightDeviceAddr nullptr; aclTensor* self nullptr; aclTensor* weight nullptr; aclTensor* gradOutput nullptr; aclTensor* gradInput nullptr; aclTensor* gradWeight nullptr; std::vectorfloat selfHostData {0, 1, 2, 3, 4, 5, 6, 7}; std::vectorfloat weightHostData {0.5, 0.5}; std::vectorfloat gradOutputHostData {1, 1, 1, 1, 1, 1, 1, 1}; std::vectorfloat gradInputHostData {0, 0, 0, 0, 0, 0, 0, 0}; std::vectorfloat gradWeightHostData {0, 0}; // 创建weight aclTensor ret CreateAclTensor(weightHostData, weightShape, weightDeviceAddr, aclDataType::ACL_FLOAT, weight); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建self aclTensor ret CreateAclTensor(selfHostData, selfShape, selfDeviceAddr, aclDataType::ACL_FLOAT, self); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建gradOutput aclTensor ret CreateAclTensor(gradOutputHostData, gradOutputShape, gradOutputDeviceAddr, aclDataType::ACL_FLOAT, gradOutput); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建gradInput aclTensor ret CreateAclTensor(gradInputHostData, gradInputShape, gradInputDeviceAddr, aclDataType::ACL_FLOAT, gradInput); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建gradWeight aclTensor ret CreateAclTensor(gradWeightHostData, gradWeightShape, gradWeightDeviceAddr, aclDataType::ACL_FLOAT, gradWeight); CHECK_RET(ret ACL_SUCCESS, return ret); // 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aclnnPreluBackward第一段接口 ret aclnnPreluBackwardGetWorkspaceSize(gradOutput, self, weight, gradInput, gradWeight, workspaceSize, executor); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnPreluBackwardGetWorkspaceSize 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aclnnPreluBackward第二段接口 ret aclnnPreluBackward(workspaceAddr, workspaceSize, executor, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnPreluBackward 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 gradInputSize GetShapeSize(gradInputShape); std::vectorfloat gradInputResultData(gradInputSize, 0); ret aclrtMemcpy(gradInputResultData.data(), gradInputResultData.size() * sizeof(gradInputResultData[0]), gradInputDeviceAddr, gradInputSize * 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 gradInputSize; i) { LOG_PRINT(gradInput[%ld] is: %f\n, i, gradInputResultData[i]); } auto gradWeightSize GetShapeSize(gradWeightShape); std::vectorfloat gradWeightResultData(gradWeightSize, 0); ret aclrtMemcpy(gradWeightResultData.data(), gradWeightResultData.size() * sizeof(gradWeightResultData[0]), gradWeightDeviceAddr, gradWeightSize * 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 gradWeightSize; i) { LOG_PRINT(gradWeight[%ld] is: %f\n, i, gradWeightResultData[i]); } // 6. 释放aclTensor和aclScalar需要根据具体API的接口定义修改 aclDestroyTensor(gradOutput); aclDestroyTensor(self); aclDestroyTensor(weight); aclDestroyTensor(gradInput); aclDestroyTensor(gradWeight); // 7. 释放device资源 需要根据具体API的接口定义修改 aclrtFree(selfDeviceAddr); aclrtFree(gradOutputDeviceAddr); aclrtFree(weightDeviceAddr); aclrtFree(gradInputDeviceAddr); aclrtFree(gradWeightDeviceAddr); if (workspaceSize 0) { aclrtFree(workspaceAddr); } aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }【免费下载链接】ops-nn本项目是CANN提供的神经网络类计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-nn创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考