aclnnGeluBackward

Applicable Products

Product	Supported
Ascend 950PR/Ascend 950DT	√
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	×
Atlas inference products	√
Atlas training products	√

Function

Description: Completes the backward propagation of aclnnGelu.
Formula:

GELU forward propagation (where x can be a scalar or Tensor):
$Gelu(x)=x \cdot \Phi(x)=x/2 \cdot [1+erf(x/\sqrt{2})]$
where the formula for erf is:
$erf(x)=\frac{2}{\sqrt \pi}\sum^{\infty}_{n=0}{\frac{(-1)^n \cdot x^{2n+1}}{n! \cdot (2n+1)}}$
The relationship between gradInput and gradOutput can be expressed as:
$gradInput = gradOutput \cdot (\frac{1}{2}+\frac{1}{2} \cdot erf(\frac{x}{\sqrt2})+\frac{x}{\sqrt{2\pi}} \cdot e^{-\frac{x^2}{2}})$
The approximate formula of Gelu is:
$Gelu(x)=0.5x(1+tanh(\sqrt{2/\pi}(x+0.044715x^3)))$

Function Prototype

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

aclnnStatus aclnnGeluBackwardGetWorkspaceSize(
  const aclTensor *gradOutput,
  const aclTensor *self,
  const aclTensor *gradInput,
  uint64_t        *workspaceSize,
  aclOpExecutor  **executor)

aclnnStatus aclnnGeluBackward(
  void             *workspace,
  uint64_t          workspace_size,
  aclOpExecutor    *executor,
  const aclrtStream stream)

aclnnGeluBackwardGetWorkspaceSize

Parameters:

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
gradOutput (aclTensor*)	Input	Weight used for gradient computation, i.e., the weight tensor multiplied with the forward output tensor to convert it into a scalar.	The shape must satisfy the broadcast relationship with the shape of the forward self. The dtype and the dtype of self must satisfy the data type deduction rules (see Deduction Relationship). Empty tensors are supported.	FLOAT, FLOAT16, BFLOAT16	ND	0-8	√
self (aclTensor*)	Input	Forward input value of Gelu.	The shape must satisfy the broadcast relationship with the shape of gradOutput The dtype must satisfy the data type deduction rules with the dtype of gradOutput (see Deduction Relationship). Empty tensors are supported.	FLOAT, FLOAT16, BFLOAT16	ND	0-8	√
gradInput (aclTensor*)	Output	Output of the backward computation, which is the gradient of the GELU forward input, i.e., the result of differentiating the input.	The dtype is consistent with the convertible data type derived from self and gradOutput (see Conversion Relationship). The shape is consistent with the shape after broadcasting gradOutput and self.	FLOAT, FLOAT16, BFLOAT16	ND	0-8	√
workspaceSize (uint64_t*)	Output	Returns the workspace size to be applied for on the Device side.	-	-	-	-	-
executor (aclOpExecutor**)	Output	Returns the op executor, which contains the operator computation flow.	-	-	-	-	-

Atlas inference products, Atlas training products: The supported data types are FLOAT and FLOAT16.

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 gradOutput, self, or gradInput is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data type or data format of gradOutput, self, or gradInput is not within the supported range.
		The dimension relationship of gradOutput, self, and gradInput does not satisfy the broadcast principle.
		The data types of gradOutput, self, and gradInput do not satisfy the data type deduction rule.

aclnnGeluBackward

Parameters:

Parameter	Input/Output	Description
workspace	Input	Workspace memory address applied for on the Device side.
workspaceSize	Input	Workspace size applied for on the Device side, obtained by the first-phase API aclnnGeluBackwardGetWorkspaceSize.
executor	Input	Operator executor that contains the operator computation process.
stream	Input	Specifies the stream for task execution.

Return value

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

Constraints

Deterministic computation:
- aclnnGeluBackward 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_gelu_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 shape_size = 1;
    for (auto i : shape) {
        shape_size *= i;
    }
    return shape_size;
}

int Init(int32_t deviceId, aclrtStream* stream)
{
    // Boilerplate: initialize resources.
    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 apply for 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 data from the host to the 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 ACL API documentation.
    // Set deviceId based on the actual device.
    int32_t deviceId = 0;
    aclrtStream stream;
    auto ret = Init(deviceId, &stream);
    CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
    // 2. Construct the input and output. Customize the construction based on the API.
    std::vector<int64_t> selfShape = {4, 2};
    std::vector<int64_t> gradOutputShape = {4, 2};
    std::vector<int64_t> gradInputShape = {4, 2};
    void* selfDeviceAddr = nullptr;
    void* gradOutputDeviceAddr = nullptr;
    void* gradInputDeviceAddr = nullptr;
    aclTensor* self = nullptr;
    aclTensor* gradOutput = nullptr;
    aclTensor* gradInput = nullptr;
    std::vector<float> selfHostData = {0, 1, 2, 3, 4, 5, 6, 7};
    std::vector<int> gradOutputHostData = {1, 1, 1, 1, 1, 1, 1, 1};
    std::vector<int> gradInputHostData = {0, 0, 0, 0, 0, 0, 0, 0};

    ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(
        gradOutputHostData, gradOutputShape, &gradOutputDeviceAddr, aclDataType::ACL_INT32, &gradOutput);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(gradInputHostData, gradInputShape, &gradInputDeviceAddr, aclDataType::ACL_FLOAT, &gradInput);
    CHECK_RET(ret == ACL_SUCCESS, return ret);

    // 3. Call the CANN operator library API. Replace with the specific API.
    uint64_t workspaceSize = 0;
    aclOpExecutor* executor;
    // Call the first-phase API of aclnnGeluBackward.
    ret = aclnnGeluBackwardGetWorkspaceSize(gradOutput, self, gradInput, &workspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGeluBackwardGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);

    // Apply for device memory based on the workspaceSize calculated 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 aclnnGeluBackward.
    ret = aclnnGeluBackward(workspaceAddr, workspaceSize, executor, stream);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGeluBackward failed. ERROR: %d\n", ret); return ret);
    // 4. (Boilerplate) Synchronously wait for the 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 the device memory to the host memory. Modify based on the specific API definition.
    auto size = GetShapeSize(gradInputShape);
    std::vector<float> resultData(size, 0);
    ret = aclrtMemcpy(
        resultData.data(), resultData.size() * sizeof(resultData[0]), gradInputDeviceAddr, size * 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 < size; i++) {
        LOG_PRINT("result[%ld] is: %f\n", i, resultData[i]);
    }
    // 6. Release aclTensor and aclScalar. Modify based on the specific API definition.
    aclDestroyTensor(gradOutput);
    aclDestroyTensor(self);
    aclDestroyTensor(gradInput);

    // 7. Release device resources. Modify based on the specific API definition.
    aclrtFree(selfDeviceAddr);
    aclrtFree(gradOutputDeviceAddr);
    aclrtFree(gradInputDeviceAddr);
    if (workspaceSize > 0) {
        aclrtFree(workspaceAddr);
    }
    aclrtDestroyStream(stream);
    aclrtResetDevice(deviceId);
    aclFinalize();
    return 0;
}