aclnnGeluMul

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:

Splits the input tensor into two tensors along the last dimension: x1 and x2. Applies the Gelu operation to x1 on the left, and multiplies the result with x2.
Formula:

Given an input tensor input whose last dimension length is 2d, the function GeluMul performs the following computation:
1. Split input into two parts:
  $x_1 = \text{input}[..., :d], \quad x_2 = \text{input}[..., d:]$
2. Apply the GELU activation function to x1. The formula for the "tanh" mode is as follows:
  $\text{GELU}(x) = 0.5 \cdot x \cdot \left( 1 + \tanh\left( \sqrt{\frac{2}{\pi}} \cdot \left( x + 0.044715 x^3 \right) \right) \right)$
  The formula for the "none" mode (erf mode) is as follows:
  $\text{GELU}(x) = 0.5 \cdot x \left( 1 + \text{erf}\left( \frac{x}{\sqrt{2}} \right) \right)$
  Therefore, the computation is as follows:
  $x_1 = \text{GELU}(x_1)$
3. The final output is the element-wise product of x1 and x2:
  $\text{out} = x_1 \times x_2$

Function Prototype

Each operator is divided into a two-phase API. You must first call the "aclnnGeluMulGetWorkspaceSize" API to obtain the input parameters and calculate the required workspace size based on the computation process, and then call the "aclnnGeluMul" API to perform the computation.

aclnnStatus aclnnGeluMulGetWorkspaceSize(
  const aclTensor *input,
  char            *approximateOptional,
  const aclTensor *out,
  uint64_t        *workspaceSize,
  aclOpExecutor  **executor)

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

aclnnGeluMulGetWorkspaceSize

Parameters:

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
input (aclTensor*)	Input	Input tensor, which is the input in the formula.	Supports empty tensors. The last dimension is an even number and less than or equal to 1024. The product of other dimensions is less than or equal to 200000.	BFLOAT16, FLOAT16, FLOAT	ND	2-8	√
approximateOptional（char*）	Input	GELU computation mode.	Only supports "none" and "tanh", corresponding to the erf mode and tanh mode of GELU respectively. Defaults to "none" when the input is a null pointer.	-	-	-	-
out (aclTensor*)	Output	Output tensor, i.e., out in the formula.	The output data type is the same as the input data type. The output shape is the same as the input shape in all dimensions except the last. The value of the last dimension is half of the last dimension value of the input shape.	BFLOAT16, FLOAT16, FLOAT	ND	2-8	√
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.	-	-	-	-	-

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 or out parameter is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data type of input is not within the supported range.

aclnnGeluMul

Parameters:

Parameter	Input/Output	Description
workspace	Input	Address of the workspace memory applied on the Device side.
workspaceSize	Input	Size of the workspace applied on the Device side, obtained through the first-phase API aclnnGeluMulGetWorkspaceSize.
executor	Input	Operator executor, which 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:
- aclnnGeluMul defaults to a deterministic implementation.
In typical scenarios, the last axis is a multiple of 16. When the last axis is not 32-byte aligned, it is recommended to use the small operator splicing logic.

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_mul.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;
}

void PrintOutResult(std::vector<int64_t> &shape, void** deviceAddr) {
  auto size = GetShapeSize(shape);
  std::vector<float> resultData(size, 0);
  auto ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]),
                         *deviceAddr, 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);
  for (int64_t i = 0; i < size; i++) {
    LOG_PRINT("mean result[%ld] is: %f\n", i, resultData[i]);
  }
}

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 memory on the device side.
  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);
  // Call aclrtMemcpy to copy data from the host side to the device memory.
  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);

  // Compute 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];
  }

  // Call the aclCreateTensor API to create an 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. (Boilerplate) Initialize the device/stream. See the ACL API manual.
  // Fill in the deviceId based on your actual device.
  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. Construct the inputs and outputs. Customize the construction based on the API interface.
  std::vector<int64_t> inputShape = {2, 4};

  std::vector<float> inputHostData = {0, 1, 2, 3, 4, 5, 6, 7};

  void* inputDeviceAddr = nullptr;

  aclTensor* input = nullptr;
  // Create the input aclTensor.
  ret = CreateAclTensor(inputHostData, inputShape, &inputDeviceAddr, aclDataType::ACL_FLOAT, &input);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  char approximate[] = "tanh";

  std::vector<int64_t> outShape = {2, 2};
  std::vector<float> outHostData(2 * 2, 1);
  aclTensor* out = nullptr;
  void* outDeviceAddr = nullptr;
  // Create the out aclTensor.
  ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  // 3. Call the CANN operator library API. Modify the API name as needed.
  uint64_t workspaceSize = 16 * 1024 * 1024;
  aclOpExecutor* executor;

  // Call the first-phase API of aclnnGeluMul.
  ret = aclnnGeluMulGetWorkspaceSize(input, approximate, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGeluMulGetWorkspaceSize 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 aclnnGeluMul.
  ret = aclnnGeluMul(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGeluMul 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, copy the result from the device memory to the host side, and modify it based on the specific API definition.
  PrintOutResult(outShape, &outDeviceAddr);


  // 6. Release the aclTensor. Modify based on the specific API definition.
  aclDestroyTensor(input);
  aclDestroyTensor(out);

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

  return 0;
}