aclnnScaledMaskedSoftmax

Applicable Products

Product	Supported
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√

Function

Description: Scales the input data x, applies a mask, and then performs Softmax to produce the output.
Formula:
$y = Softmax((scale * x) * mask, dim = -1)$ $Softmax(X_i) ={e^{X_i - max(X, dim=-1)} \over \sum{e^{X_i - max(X, dim=-1)}}}$

Function Prototype

Each operator is divided into a two-phase API. You must first call the "aclnnScaledMaskedSoftmaxGetWorkspaceSize" API to obtain the workspace size required for computation and the executor that contains the operator computation flow, and then call the "aclnnScaledMaskedSoftmax" API to perform the computation.

aclnnStatus aclnnScaledMaskedSoftmaxGetWorkspaceSize(
    const aclTensor* x,
    const aclTensor* mask,
    double scale,
    bool fixTriuMask,
    aclTensor*       y,
    uint64_t*        workspaceSize,
    aclOpExecutor**  executor)

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

aclnnScaledMaskedSoftmaxGetWorkspaceSize

Parameters:

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
x	Input	Input data, i.e., x in the formula.	The shape must satisfy the broadcast relationship with mask.	FLOAT, FLOAT16, BFLOAT16	ND	[B,N,S1,S2]	√
mask	Input	Mask to be applied to x, i.e., mask in the formula.	The shape must satisfy the broadcast relationship with x.	BOOL	ND	[B,N,S1,S2], [B,1,S1,S2], [1,N,S1,S2], [1,1,S1,S2]	√
scale	Input	Magnitude of data scaling, i.e., scale in the formula.	-	DOUBLE	-	-	-
fixTriuMask	Input.	Indicates whether to generate an upper triangular mask tensor within the operator.	Only false is supported.	BOOL	-	-	-
y	Output	Softmax result of scaledMaskedX, corresponding to y in the formula.	The data type and shape must be consistent with x.	FLOAT, FLOAT16, BFLOAT16	ND	[B,N,S1,S2]	×
workspaceSize	output	Returns the workspace size to be applied for on the Device side.	-	-	-	-	-
executor	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 Value	Error Code	Description
ACLNN_ERR_PARAM_NULLPTR	161001	x and y are null pointers.
ACLNN_ERR_PARAM_NULLPTR	161001	When fixedTriuMask is false, mask is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data types of x, mask, and y are not within the supported range.
		The shape of x, mask, or y is not 4-dimensional.
		The fourth dimension of x is greater than 4096 or equals 0.
		The shapes of x and y are different.
		mask cannot be broadcast to the shape of x.

aclnnScaledMaskedSoftmax

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 from the first-phase API aclnnScaledMaskedSoftmaxGetWorkspaceSize.
executor	Input	Operator executor that 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:
- aclnnScaledMaskedSoftmax defaults to a deterministic implementation.
The fourth dimension of x must be within $(0,4096]$ . The first two dimensions of mask may differ from those of x, but must satisfy the broadcast relationship.

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_scaled_masked_softmax.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;
}

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: resource initialization.
  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);

  // Call aclrtMemcpy to copy host-side data to device-side 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);

  // 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];
  }

  // Call aclCreateTensor 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 input and output. Customize the construction based on the API.
  // input
  std::vector<float> xHostData = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
  std::vector<char> maskHostData = {1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0};
  std::vector<float> yHostData(16, 0);
  std::vector<int64_t> xShape = {2, 2, 2, 2};
  std::vector<int64_t> maskShape = {2, 2, 2, 2};
  std::vector<int64_t> yShape = {2, 2, 2, 2};
  void *xDeviceAddr = nullptr;
  void *maskDeviceAddr = nullptr;
  void *yDeviceAddr = nullptr;
  aclTensor *x = nullptr;
  aclTensor *mask = nullptr;
  aclTensor *y = nullptr;

  ret = CreateAclTensor(xHostData, xShape, &xDeviceAddr, aclDataType::ACL_FLOAT, &x);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(maskHostData, maskShape, &maskDeviceAddr, aclDataType::ACL_BOOL, &mask);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(yHostData, yShape, &yDeviceAddr, aclDataType::ACL_FLOAT, &y);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  // attr
  float scale = 1.0f;
  bool triuMask = false;

  uint64_t workspaceSize = 0;
  aclOpExecutor *executor;

  // 3. Call the CANN operator library API. Replace with the specific API name.
  // aclnnScaledMaskedSoftmax
  ret = aclnnScaledMaskedSoftmaxGetWorkspaceSize(x, mask, scale, triuMask, y, &workspaceSize, &executor);
  CHECK_RET(
      ret == ACL_SUCCESS,
      LOG_PRINT("aclnnScaledMaskedSoftmaxGetWorkspaceSize failed. ERROR: %d\n", ret);
      return ret);

  // Allocate 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);
  }

  // aclnnScaledMaskedSoftmax
  ret = aclnnScaledMaskedSoftmax(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS,
            LOG_PRINT("aclnnScaledMaskedSoftmax failed. ERROR: %d\n", ret);
            return ret);

  // 4. (Boilerplate) Wait for the task execution to complete synchronously.
  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.
  PrintOutResult(yShape, &yDeviceAddr);

  auto size = GetShapeSize(yShape);
  std::vector<float> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), yDeviceAddr,
                    size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST);

  // 6. Release aclTensor and aclScalar. Modify based on the specific API definition.
  aclDestroyTensor(x);
  aclDestroyTensor(mask);
  aclDestroyTensor(y);

  // 7. Release device resources.
  aclrtFree(xDeviceAddr);
  aclrtFree(maskDeviceAddr);
  aclrtFree(yDeviceAddr);

  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();

  return 0;
}