aclnnMultinomialTensor

Applicable Products

  • Ascend 950PR&950DT products: supported
  • Atlas A3 products: supported
  • Atlas A2 products: supported
  • Atlas 200I/500 A2 inference products: not supported
  • Atlas inference products: not supported
  • Atlas training products: not supported

Function

Draws numsamples samples from the input tensor based on the probability distribution of each object, and stores the indices of these samples in the output tensor.

Function Prototype

Each operator uses a two-phase API. Call the "aclnnMultinomialTensorGetWorkspaceSize" API first to obtain the workspace size required for the computation and the executor that contains the operator's computation process, and then call the "aclnnMultinomialTensor" API to execute the computation.

aclnnStatus aclnnMultinomialTensorGetWorkspaceSize(
  const aclTensor*      self,
  int64_t               numsamples,
  bool                  replacement,
  const aclTensor*      seedTensor,
  const aclTensor*      offsetTensor,
  int64_t               offset,
  aclTensor*            out,
  uint64_t*             workspaceSize,
  aclOpExecutor**       executor)
aclnnStatus aclnnMultinomialTensor(
  void*                 workspace,
  uint64_t              workspaceSize,
  aclOpExecutor*        executor,
  aclrtStream           stream)

aclnnMultinomialTensorGetWorkspaceSize

  • Parameters:

    Parameter Input/Output Description Instruction Data Type Data Format Dimensions (shape) non-contiguous tensor
    self (aclTensor*) Input Input tensor, indicating the probability of the distribution for each object. The values in self must be greater than or equal to 0. BFLOAT16, FLOAT16, FLOAT, DOUBLE ND (N, C) or (C), and the dimensions of self must be the same as those of out. √
    numsamples (int64_t) Input Number of samples drawn from each multinomial distribution. numsamples is a positive integer. When replacement is false, numsamples must not be greater than C. INT64 - - -
    replacement (bool) Input Determines whether elements are drawn with replacement during sampling. - BOOL - - -
    seedTensor (aclTensor*) Input Sets the seed value of the random number generator, which affects the sequence of generated random numbers. - INT64 ND [1]. √
    offsetTensor (aclTensor*) Input The accumulation result with the scalar offset is used as the offset of the random number operator. The offset of random numbers affects the position of the generated random number sequence. After the offset is set, the generated random number sequence starts from the specified position. - INT64 ND [1]. √
    offset (int64_t) Input Serves as the accumulator for offsetTensor. - INT64 - - -
    out (aclTensor*) Output Output tensor. - INT64 ND The value is (N, numsamples) or (numsamples). The dimensions of self and out must be consistent. √
    workspaceSize (uint64_t*) Output Returns the size of the workspace to be allocated on the Device. - - - - -
    executor (aclOpExecutor**) Output Returns the op executor, which includes the computation flow of the operator. - - - - -
    • Ascend 950PR&950DT products: The DOUBLE data type is not supported.
  • Return value

    aclnnStatus: return status code. For details, see aclnn Return Codes.

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

    Error Code Error Code Description
    ACLNN_ERR_PARAM_NULLPTR 161001 The input self or out is a null pointer.
    ACLNN_ERR_PARAM_INVALID 161002 The data type of self is not supported.
    The data type of out is not INT64.
    The data type of seedTensor or offsetTensor is not INT64.
    The dimensions of self are not 1D or 2D.
    The dimensions of self and out are inconsistent.
    The value of numsamples is less than or equal to 0.
    replacement is false and numsamples is greater than the size of the last dimension of self.
    The size of the last dimension of self cannot exceed 2^24.
    The size of the last dimension of out must be the same as the value of numsamples.
    When self is a 2D tensor, the first dimension of out must be equal to the size of the first dimension of self.

aclnnMultinomialTensor

  • Parameters:

    Parameter Input/Output Description
    workspace Input Workspace memory address allocated on the Device side.
    workspaceSize Input Workspace size allocated on the Device side, obtained by the first-phase API aclnnMultinomialTensorGetWorkspaceSize.
    executor Input Executor of the op, which contains the computation process of the operator.
    stream Input Stream on which tasks are executed.
  • Return value

    aclnnStatus: return status code. For details, see aclnn Return Codes.

Constraints

  • deterministic computation:
    • aclnnMultinomialTensor defaults to a deterministic implementation.

Example

The following sample code is for reference only. For details about compilation and execution, see Compile and Run Samples.

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_multinomial.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: 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 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);
  // 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 the 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 device/stream. For details, see the acl API manual.
  // Set deviceId based on the 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. Custom construction is required based on the API interface.
  std::vector<int64_t> selfShape = {8};
  std::vector<int64_t> outShape = {4};
  std::vector<int64_t> seedShape = {1};
  std::vector<int64_t> offsetShape = {1};
  void* selfDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* self = nullptr;
  aclTensor* out = nullptr;
  void* seedDeviceAddr = nullptr;
  aclTensor* seed = nullptr;
  void* offsetDeviceAddr = nullptr;
  aclTensor* offset = nullptr;
  int64_t offset2 = 102;
  std::vector<float> selfHostData = {0, 10, 3, 0, 1, 2, 1, 1};
  std::vector<int64_t> outHostData = {2, 0, 1, 2};
  std::vector<int64_t> seedHostData = {0};
  std::vector<int64_t> offsetHostData = {392};
  int64_t numsamples = 4;
  bool replacement = false;

  // Create the self aclTensor.
  ret = CreateAclTensor(selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the seed aclTensor.
  ret = CreateAclTensor(seedHostData, seedShape, &seedDeviceAddr, aclDataType::ACL_INT64, &seed);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the offset aclTensor.
  ret = CreateAclTensor(offsetHostData, offsetShape, &offsetDeviceAddr, aclDataType::ACL_INT64, &offset);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_INT64, &out);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  // 3. Call the CANN operator library API. Change it to the specific API name.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // Call the first-phase API of aclnnMultinomialTensor.
  ret = aclnnMultinomialTensorGetWorkspaceSize(self, numsamples, replacement, seed, offset, offset2, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMultinomialTensorGetWorkspaceSize 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);
  }
  // Call the second-phase API of aclnnMultinomialTensor.
  ret = aclnnMultinomialTensor(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMultinomialTensor failed. ERROR: %d\n", ret); return ret);

  // 4. (Boilerplate) Synchronize and wait for the task execution to finish.
  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 side to the host side. Modify it based on the specific API definition.
  auto size = GetShapeSize(outShape);
  std::vector<int64_t> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), outDeviceAddr,
                    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: %ld\n", i, resultData[i]);
  }

  // 6. Release aclTensor and aclScalar. Modify it based on the specific API definition.
  aclDestroyTensor(self);
  aclDestroyTensor(out);
  aclDestroyTensor(seed);
  aclDestroyTensor(offset);

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

  return 0;
}