aclnnApplyAdamW

📄 View Source Code

Applicable Products

Function

• Description: Implements the AdamW optimizer function.

• Formula:

  $$g_t=\begin{cases}-g_t & \text{ if } maximize= true\\ g_t & \text{ if } maximize=false \end{cases}$$ $$m_{t}=\beta_{1} m_{t-1}+\left(1-\beta_{1}\right) g_{t} \\$$ $$v_{t}=\beta_{2} v_{t-1}+\left(1-\beta_{2}\right) g_{t}^{2}$$ $$\beta_{1}^{t}=\beta_{1}^{t-1}\times\beta_{1}$$ $$\beta_{2}^{t}=\beta_{2}^{t-1}\times\beta_{2}$$ $$v_t=\begin{cases}\max(maxGradNorm, v_t) & \text{ if } amsgrad = true\\ v_t & \text{ if } amsgrad = false \end{cases}$$ $$\hat{m}_{t}=\frac{m_{t}}{1-\beta_{1}^{t}} \\$$ $$\hat{v}_{t}=\frac{v_{t}}{1-\beta_{2}^{t}} \\$$ $$\theta_{t+1}=\theta_{t}-\frac{\eta}{\sqrt{\hat{v}_{t}}+\epsilon} \hat{m}_{t}-\eta \cdot \lambda \cdot \theta_{t-1}$$

Function Prototype

Each operator is divided into a two-phase API. The "aclnnApplyAdamWGetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnApplyAdamW" API is called to perform the computation.

aclnnStatus aclnnApplyAdamWGetWorkspaceSize(
    aclTensor*       varRef,
    aclTensor*       mRef,
    aclTensor*       vRef,
    const aclTensor* beta1Power,
    const aclTensor* beta2Power,
    const aclTensor* lr,
    const aclTensor* weightDecay,
    const aclTensor* beta1,
    const aclTensor* beta2,
    const aclTensor* eps,
    const aclTensor* grad,
    const aclTensor* maxGradNormOptional,
    bool             amsgrad,
    bool             maximize,
    uint64_t*        workspaceSize,
    aclOpExecutor**  executor)

aclnnStatus aclnnApplyAdamW(
    void            *workspace,
    uint64_t         workspaceSize,
    aclOpExecutor   *executor,
    aclrtStream      stream)

aclnnApplyAdamWGetWorkspaceSize

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  varRef (aclTensor*)	Input/Output	The weight input to be computed, which is also the output, i.e., θ in the formula.		FLOAT16, BFLOAT16, FLOAT32	ND	1-8	√
  mRef (aclTensor*)	Input/Output	The m parameter in the AdamW optimizer, i.e., m in the formula.	The shape must be consistent with the input varRef.	Must be consistent with varRef.	ND	1-8	√
  vRef (aclTensor*)	Input/Output	The v parameter in the AdamW optimizer, i.e., v in the formula.	The shape must be consistent with the input varRef.	Consistent with varRef.	ND	1-8	√
  beta1Power (aclTensor*)	Input	The beta1^(t-1) parameter.	The shape must be [1].	Must be consistent with varRef.	ND	1	√
  beta2Power (aclTensor*)	Input	The beta2^(t-1) parameter.	The shape must be [1].	Must be consistent with varRef.	ND	1	√
  lr (aclTensor*)	Input	Learning rate, denoted as η in the formula, typically 1e-3, 1e-6, or 1e-9.	The shape must be [1].	Must be consistent with varRef.	ND	1	√
  weightDecay (aclTensor*)	Input	Weight decay coefficient.	The shape must be [1].	Must be consistent with varRef.	ND	1	√
  beta1 (aclTensor*)	Input	beta1 parameter.	The shape must be [1].	The data type must be consistent with varRef.	ND	1	√
  beta2 (aclTensor*)	Input	The beta2 parameter.	The shape must be [1].	Must be consistent with varRef.	ND	1	√
  eps (aclTensor*)	Input	Prevents division by zero.	The shape must be [1].	Must be consistent with varRef.	ND	1	√
  grad (aclTensor*)	Input	Gradient data, i.e., g in the formula.	The shape must be consistent with the input varRef.	Consistent with varRef.	ND	1-8	√
  maxGradNormOptional (aclTensor*)	Input	If amsgrad=true, saves the maximum of maxGradNorm and v in vRef. If amsgrad=false, performs no additional computation on the value in vRef. This parameter is mandatory when the amsgrad parameter is true, and optional when the amsgrad parameter is false.	The shape must be consistent with the input varRef.	Must be consistent with varRef.	ND	1-8	√
  amsgrad (bool)	Input	Whether to use the maxGradNormOptional variable.	-	bool	-	-	-
  maximize (bool)	Input	Specifies whether to negate the gradient grad and apply the gradient ascent direction to optimize weights to maximize the loss function.	-	bool	-	-	-
  workspaceSize (uint64_t*)	Output	Returns the workspace size to be allocated on the Device side.	-	-	-	-	-
  executor (aclOpExecutor**)	Output	Returns the op executor, which contains the operator computation process.	-	-	-	-	-

• Return value

  aclnnStatus: return code. For details, see aclnn Return Code.

  The first-phase API performs input parameter validation and returns an error in the following scenarios:

  Return Code	Error Code	Description
  ACLNN_ERR_PARAM_NULLPTR	161001	The input computation parameter passed in is a null pointer.
  		When amsgrad is true, maxGradNormOptional is a null pointer.
  ACLNN_ERR_PARAM_INVALID	161002	The data type of the input computation input is not within the supported range.
  		The data types of the input computation inputs are inconsistent.
  		The shapes of the input computation inputs are inconsistent.
  		When amsgrad is true, the data type or shape of maxGradNormOptional and varRef are inconsistent.
  		The shape size of beta1Power, beta2Power, lr, weightDecay, beta1, beta2, or eps is not 1.

aclnnApplyAdamW

• Parameters:

  Parameter	Input/Output	Description
  workspace	Input	Address of the workspace memory applied for on the Device side.
  workspaceSize	Input	Size of the workspace applied for on the Device side, obtained through the first-phase API aclnnApplyAdamWGetWorkspaceSize.
  executor	Input	Operator executor, which contains the operator computation flow.
  stream	Input	Specifies the stream for task execution.

• Return value

  aclnnStatus: return code. For details, see aclnn Return Code.

Constraints

• Deterministic computation:

  • aclnnApplyAdamW defaults to a deterministic implementation.

Example

The following provides example code for reference only. For details about compilation and running, see Compile and Run Samples.

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_apply_adam_w.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) {
  // Boilerplate: initialize AscendCL.
  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);
  // Call aclrtMalloc to allocate device-side memory.
  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);

  // Copy host-side data to device memory by calling aclrtMemcpy.
  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);

  // Calculate the strides of a contiguous tensor.
  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];
  }

  // Create an aclTensor by calling aclCreateTensor.
  *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
                            shape.data(), shape.size(), *deviceAddr);
  return 0;
}

int main() {
  // 1. (Boilerplate) Initialize the device/stream. See the AscendCL API list.
  // Set deviceId based on the actual device.
  int32_t deviceId = 0;
  aclrtStream stream;
  auto ret = Init(deviceId, &stream);
  // Handle the check based on your needs.
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);

  // 2. Construct the input and output. Customize the construction based on the API interface.
  std::vector<int64_t> varShape = {2, 2};
  std::vector<int64_t> mShape = {2, 2};
  std::vector<int64_t> vShape = {2, 2};
  std::vector<int64_t> beta1PowerShape = {1};
  std::vector<int64_t> beta2PowerShape = {1};
  std::vector<int64_t> lrShape = {1};
  std::vector<int64_t> weightDecayShape = {1};
  std::vector<int64_t> beta1Shape = {1};
  std::vector<int64_t> beta2Shape = {1};
  std::vector<int64_t> epsShape = {1};
  std::vector<int64_t> gradShape = {2, 2};
  std::vector<int64_t> maxgradShape = {2, 2};
  void* varDeviceAddr = nullptr;
  void* mDeviceAddr = nullptr;
  void* vDeviceAddr = nullptr;
  void* beta1PowerDeviceAddr = nullptr;
  void* beta2PowerDeviceAddr = nullptr;
  void* lrDeviceAddr = nullptr;
  void* weightDecayDeviceAddr = nullptr;
  void* beta1DeviceAddr = nullptr;
  void* beta2DeviceAddr = nullptr;
  void* epsDeviceAddr = nullptr;
  void* gradDeviceAddr = nullptr;
  void* maxgradDeviceAddr = nullptr;
  aclTensor* var = nullptr;
  aclTensor* m = nullptr;
  aclTensor* v = nullptr;
  aclTensor* beta1Power = nullptr;
  aclTensor* beta2Power = nullptr;
  aclTensor* lr = nullptr;
  aclTensor* weightDecay = nullptr;
  aclTensor* beta1 = nullptr;
  aclTensor* beta2 = nullptr;
  aclTensor* eps = nullptr;
  aclTensor* grad = nullptr;
  aclTensor* maxgrad = nullptr;
  std::vector<float> varHostData = {0, 1, 2, 3};
  std::vector<float> mHostData = {0, 1, 2, 3};
  std::vector<float> vHostData = {0, 1, 2, 3};
  std::vector<float> beta1PowerHostData = {0.431};
  std::vector<float> beta2PowerHostData = {0.992};
  std::vector<float> lrHostData = {0.001};
  std::vector<float> weightDecayHostData = {0.01};
  std::vector<float> beta1HostData = {0.9};
  std::vector<float> beta2HostData = {0.999};
  std::vector<float> epsHostData = {1e-8};
  std::vector<float> gradHostData = {0, 1, 2, 3};
  std::vector<float> maxgradHostData = {0, 1, 2, 3};
  bool amsgrad = true;
  bool maximize = true;
  // Create the var aclTensor.
  ret = CreateAclTensor(varHostData, varShape, &varDeviceAddr, aclDataType::ACL_FLOAT, &var);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the m aclTensor.
  ret = CreateAclTensor(mHostData, mShape, &mDeviceAddr, aclDataType::ACL_FLOAT, &m);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the v aclTensor.
  ret = CreateAclTensor(vHostData, vShape, &vDeviceAddr, aclDataType::ACL_FLOAT, &v);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create beta1Power aclTensor.
  ret = CreateAclTensor(beta1PowerHostData, beta1PowerShape, &beta1PowerDeviceAddr, aclDataType::ACL_FLOAT, &beta1Power);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create beta2Power aclTensor.
  ret = CreateAclTensor(beta2PowerHostData, beta2PowerShape, &beta2PowerDeviceAddr, aclDataType::ACL_FLOAT, &beta2Power);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create lr aclTensor.
  ret = CreateAclTensor(lrHostData, lrShape, &lrDeviceAddr, aclDataType::ACL_FLOAT, &lr);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create weightDecay aclTensor.
  ret = CreateAclTensor(weightDecayHostData, weightDecayShape, &weightDecayDeviceAddr, aclDataType::ACL_FLOAT, &weightDecay);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create beta1 aclTensor.
  ret = CreateAclTensor(beta1HostData, beta1Shape, &beta1DeviceAddr, aclDataType::ACL_FLOAT, &beta1);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the beta2 aclTensor.
  ret = CreateAclTensor(beta2HostData, beta2Shape, &beta2DeviceAddr, aclDataType::ACL_FLOAT, &beta2);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the eps aclTensor.
  ret = CreateAclTensor(epsHostData, epsShape, &epsDeviceAddr, aclDataType::ACL_FLOAT, &eps);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the grad aclTensor.
  ret = CreateAclTensor(gradHostData, gradShape, &gradDeviceAddr, aclDataType::ACL_FLOAT, &grad);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the maxgrad aclTensor.
  ret = CreateAclTensor(maxgradHostData, maxgradShape, &maxgradDeviceAddr, aclDataType::ACL_FLOAT, &maxgrad);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  // 3. Call the CANN operator library API. Replace with the actual API name.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // Call the first-phase API of aclnnApplyAdamW.
  ret = aclnnApplyAdamWGetWorkspaceSize(var, m, v, beta1Power, beta2Power, lr, weightDecay, beta1, beta2, eps, grad, maxgrad, amsgrad, maximize, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnApplyAdamWGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
  // Allocate device memory based on the workspaceSize computed by the first-phase API.
  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);
  }
  // Call the second-phase API of aclnnApplyAdamW.
  ret = aclnnApplyAdamW(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnApplyAdamW failed. ERROR: %d\n", ret); return ret);

  // 4. (Boilerplate) Synchronize and wait for task execution to complete.
  ret = aclrtSynchronizeStream(stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);

  // 5. Obtain the output value and copy the result from device memory to host memory. Modify this based on the specific API definition.
  auto size = GetShapeSize(varShape);
  std::vector<float> 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]);
  }

  // 6. Release aclTensor. Modify based on the specific API definition.
  aclDestroyTensor(var);
  aclDestroyTensor(m);
  aclDestroyTensor(v);
  aclDestroyTensor(beta1Power);
  aclDestroyTensor(beta2Power);
  aclDestroyTensor(lr);
  aclDestroyTensor(weightDecay);
  aclDestroyTensor(beta1);
  aclDestroyTensor(beta2);
  aclDestroyTensor(eps);
  aclDestroyTensor(grad);
  aclDestroyTensor(maxgrad);

  // 7. Release device resources.
  aclrtFree(varDeviceAddr);
  aclrtFree(mDeviceAddr);
  aclrtFree(vDeviceAddr);
  aclrtFree(beta1PowerDeviceAddr);
  aclrtFree(beta2PowerDeviceAddr);
  aclrtFree(lrDeviceAddr);
  aclrtFree(weightDecayDeviceAddr);
  aclrtFree(beta1DeviceAddr);
  aclrtFree(beta2DeviceAddr);
  aclrtFree(epsDeviceAddr);
  aclrtFree(gradDeviceAddr);
  aclrtFree(maxgradDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}