aclnnApplyAdam【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×功能说明接口功能实现Adam优化器功能。计算公式$$ lr_t lr \times \frac{\sqrt{1 - \beta_2^t}}{1 - \beta_1^t} $$$$ m_t \beta_1 \times m_{t-1} (1 - \beta_1) \times g_t $$$$ v_t \beta_2 \times v_{t-1} (1 - \beta_2) \times g_t^2 $$若 use_nesterov true: $$ var_t var_{t-1} - lr_t \times \frac{\beta_1 \times m_t (1 - \beta_1) \times g_t}{\sqrt{v_t} \epsilon} $$若 use_nesterov false: $$ var_t var_{t-1} - lr_t \times \frac{m_t}{\sqrt{v_t} \epsilon} $$函数原型每个算子分为两段式接口必须先调用aclnnApplyAdamGetWorkspaceSize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aclnnApplyAdam接口执行计算。aclnnStatus aclnnApplyAdamGetWorkspaceSize( aclTensor *varRef, const aclTensor *m, const aclTensor *v, const aclTensor *beta1Power, const aclTensor *beta2Power, const aclTensor *lr, const aclTensor *beta1, const aclTensor *beta2, const aclTensor *epsilon, const aclTensor *grad, bool useLocking, bool useNesterov, uint64_t *workspaceSize, aclOpExecutor **executor)aclnnStatus aclnnApplyAdam( void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream)aclnnApplyAdamGetWorkspaceSize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TensorvarRefaclTensor*输入/输出待计算的权重输入同时也是输出公式中的var。FLOAT16、BFLOAT16、FLOAT32ND1-8√maclTensor*输入Adam优化器中m参数公式中的m。shape要求与输入varRef保持一致。与varRef保持一致ND1-8√vaclTensor*输入Adam优化器中v参数公式中的v。shape要求与输入varRef保持一致。与varRef保持一致ND1-8√beta1PoweraclTensor*输入beta1的t次幂。shape要求为[1]。FLOAT32ND1√beta2PoweraclTensor*输入beta2的t次幂。shape要求为[1]。FLOAT32ND1√lraclTensor*输入学习率公式中的lr。shape要求为[1]。FLOAT32ND1√beta1aclTensor*输入beta1参数。shape要求为[1]。FLOAT32ND1√beta2aclTensor*输入beta2参数。shape要求为[1]。FLOAT32ND1√epsilonaclTensor*输入防止除数为0。shape要求为[1]。FLOAT32ND1√gradaclTensor*输入梯度数据公式中的gshape要求与输入varRef保持一致。与varRef保持一致ND1-8√useLockingbool输入是否使用锁机制。-bool---useNesterovbool输入是否使用Nesterov momentum。-bool---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。-----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码错误码描述ACLNN_ERR_PARAM_NULLPTR161001传入的计算输入参数是空指针时。ACLNN_ERR_PARAM_INVALID161002传入的计算输入的数据类型不在支持的范围内时。传入的计算输入的数据类型不一致时。传入的计算输入的shape不一致时。beta1Power、beta2Power、lr、beta1、beta2、epsilon的数据类型不为FLOAT32时。beta1Power、beta2Power、lr、beta1、beta2、epsilon的shape大小不为1时。aclnnApplyAdam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aclnnApplyAdamGetWorkspaceSize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aclnnApplyAdam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_apply_adam.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 shapeSize 1; for (auto i : shape) { shapeSize * i; } return shapeSize; } int Init(int32_t deviceId, aclrtStream* stream) { 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); 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); 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); 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]; } *tensor aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), *deviceAddr); return 0; } int main() { 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); std::vectorint64_t varShape {2, 2}; std::vectorint64_t mShape {2, 2}; std::vectorint64_t vShape {2, 2}; std::vectorint64_t beta1PowerShape {1}; std::vectorint64_t beta2PowerShape {1}; std::vectorint64_t lrShape {1}; std::vectorint64_t beta1Shape {1}; std::vectorint64_t beta2Shape {1}; std::vectorint64_t epsilonShape {1}; std::vectorint64_t gradShape {2, 2}; void* varDeviceAddr nullptr; void* mDeviceAddr nullptr; void* vDeviceAddr nullptr; void* beta1PowerDeviceAddr nullptr; void* beta2PowerDeviceAddr nullptr; void* lrDeviceAddr nullptr; void* beta1DeviceAddr nullptr; void* beta2DeviceAddr nullptr; void* epsilonDeviceAddr nullptr; void* gradDeviceAddr nullptr; aclTensor* var nullptr; aclTensor* m nullptr; aclTensor* v nullptr; aclTensor* beta1Power nullptr; aclTensor* beta2Power nullptr; aclTensor* lr nullptr; aclTensor* beta1 nullptr; aclTensor* beta2 nullptr; aclTensor* epsilon nullptr; aclTensor* grad nullptr; std::vectorfloat varHostData {0, 1, 2, 3}; std::vectorfloat mHostData {0, 1, 2, 3}; std::vectorfloat vHostData {0, 1, 2, 3}; std::vectorfloat beta1PowerHostData {0.431}; std::vectorfloat beta2PowerHostData {0.992}; std::vectorfloat lrHostData {0.001}; std::vectorfloat beta1HostData {0.9}; std::vectorfloat beta2HostData {0.999}; std::vectorfloat epsilonHostData {1e-8}; std::vectorfloat gradHostData {0, 1, 2, 3}; bool useLocking false; bool useNesterov false; ret CreateAclTensor(varHostData, varShape, varDeviceAddr, aclDataType::ACL_FLOAT, var); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(mHostData, mShape, mDeviceAddr, aclDataType::ACL_FLOAT, m); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(vHostData, vShape, vDeviceAddr, aclDataType::ACL_FLOAT, v); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(beta1PowerHostData, beta1PowerShape, beta1PowerDeviceAddr, aclDataType::ACL_FLOAT, beta1Power); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(beta2PowerHostData, beta2PowerShape, beta2PowerDeviceAddr, aclDataType::ACL_FLOAT, beta2Power); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(lrHostData, lrShape, lrDeviceAddr, aclDataType::ACL_FLOAT, lr); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(beta1HostData, beta1Shape, beta1DeviceAddr, aclDataType::ACL_FLOAT, beta1); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(beta2HostData, beta2Shape, beta2DeviceAddr, aclDataType::ACL_FLOAT, beta2); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(epsilonHostData, epsilonShape, epsilonDeviceAddr, aclDataType::ACL_FLOAT, epsilon); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(gradHostData, gradShape, gradDeviceAddr, aclDataType::ACL_FLOAT, grad); CHECK_RET(ret ACL_SUCCESS, return ret); uint64_t workspaceSize 0; aclOpExecutor* executor; ret aclnnApplyAdamGetWorkspaceSize(var, m, v, beta1Power, beta2Power, lr, beta1, beta2, epsilon, grad, useLocking, useNesterov, workspaceSize, executor); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnApplyAdamGetWorkspaceSize failed. ERROR: %d\n, ret); return ret); 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); } ret aclnnApplyAdam(workspaceAddr, workspaceSize, executor, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnApplyAdam failed. ERROR: %d\n, ret); return ret); ret aclrtSynchronizeStream(stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtSynchronizeStream failed. ERROR: %d\n, ret); return ret); auto size GetShapeSize(varShape); std::vectorfloat resultData(size, 0); ret aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), varDeviceAddr, 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]); } aclDestroyTensor(var); aclDestroyTensor(m); aclDestroyTensor(v); aclDestroyTensor(beta1Power); aclDestroyTensor(beta2Power); aclDestroyTensor(lr); aclDestroyTensor(beta1); aclDestroyTensor(beta2); aclDestroyTensor(epsilon); aclDestroyTensor(grad); aclrtFree(varDeviceAddr); aclrtFree(mDeviceAddr); aclrtFree(vDeviceAddr); aclrtFree(beta1PowerDeviceAddr); aclrtFree(beta2PowerDeviceAddr); aclrtFree(lrDeviceAddr); aclrtFree(beta1DeviceAddr); aclrtFree(beta2DeviceAddr); aclrtFree(epsilonDeviceAddr); aclrtFree(gradDeviceAddr); 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