aclnnApplyTopKTopP

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: Performs top-k and top-p sampling filtering on the original input logits.
Formula:
- Sorts the input logits in ascending order along the last axis to obtain the corresponding sorted results sortedValue and sortedIndices.
$sortedValue, sortedIndices = sort(logits, dim=-1, descending=false, stable=true)$
- Compute the retention threshold (the k-th largest value).
$topKValue[b][v] = sortedValue[b][sortedValue.size(1) - k[b]]$
- Generate the mask for top-k filtering.
$topKMask = sortedValue < topKValue$
- Set values below the threshold to -Inf using topKMask.
$sortedValue[b][v] = \begin{cases} -Inf & \text{topKMask[b][v]=true}\\ sortedValue[b][v] & \text{topKMask[b][v]=false} \end{cases}$
- Converts the top-k filtered data into a probability distribution along the last axis using softmax.
$probsValue = softmax(sortedValue, dim=-1)$
- Computes the cumulative probability along the last axis (accumulating from the smallest probability).
$probsSum = cumsum(probsValue, dim=-1)$
- Generate the top-p mask. Positions where the cumulative probability is less than or equal to 1-p need to be filtered out, and ensure that at least one element is retained per batch.
$topPMask[b][v] = probsSum[b][v] <= 1-p[b]$ $topPMask[b][-1] = false$
- Set the values below the threshold to -Inf through topPMask.
$sortedValue[b][v] = \begin{cases} -Inf & \text{topPMask[b][v]=true}\\ sortedValue[b][v] & \text{topPMask[b][v]=false} \end{cases}$
- Restore the filtered results to the original order based on sortedIndices.
$out[b][v] = sortedValue[b][sortedIndices[b][v]]$ where $0 \le b \lt logits.size(0), 0 \le v \lt logits.size(1)$ .

Function Prototype

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

aclnnStatus aclnnApplyTopKTopPGetWorkspaceSize(
  const aclTensor* logits,
  const aclTensor* p,
  const aclTensor* k,
  aclTensor*       out,
  uint64_t*        workspaceSize,
  aclOpExecutor**  executor)

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

aclnnApplyTopKTopPGetWorkspaceSize

Parameters:

Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
logits	Input	Represents the data to be processed, i.e., logits in the formula.	-	BFLOAT16, FLOAT16, FLOAT32	ND	2	√
p	Input	Indicates the threshold for top-p, i.e., p in the formula.	The value range is [0, 1]. The data type must be consistent with logits. The shape must be consistent with logits.size(0).	BFLOAT16, FLOAT16, FLOAT32	ND	1	√
k	Input	Represents the top-k threshold, i.e., k in the formula.	The value range is [1, logits.size(1)]. The shape must be consistent with logits.size(0).	INT32	ND	1	√
out	Output	The filtered data, which is "out" in the formula.	The data type must be consistent with logits. The shape must be consistent with logits.	BFLOAT16, FLOAT16, FLOAT32	ND	2	√
workspaceSize	Output	Returns the workspace size to be requested on the Device side.	-	-	-	-	-
executor	Output	Returns the operator 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 passed logits or out is a null pointer, or both p and k are null pointers.
ACLNN_ERR_PARAM_INVALID	161002	The data type of logits, p, k, or out is not within the supported range.
		The data types of logits, p, and out do not match.
		The shapes of logits, p, k, and out do not match.

aclnnApplyTopKTopP

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 aclnnApplyTopKTopPGetWorkspaceSize.
executor	Input	Operator executor that contains the operator computation flow.
stream	Input	Specifies the stream that executes the task.

Return value

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

Constraints

Deterministic computation:
- aclnnApplyTopKTopP 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_top_k_top_p.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 for 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 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 data from the host 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.
  // 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 input and output. Customize the construction based on the API interface.
  std::vector<int64_t> logitsShape = {3, 4};
  std::vector<int64_t> pShape = {3};
  std::vector<int64_t> kShape = {3};
  std::vector<int64_t> outShape = {3, 4};
  void* logitsDeviceAddr = nullptr;
  void* pDeviceAddr = nullptr;
  void* kDeviceAddr = nullptr;
  void* outDeviceAddr = nullptr;
  aclTensor* logits = nullptr;
  aclTensor* p = nullptr;
  aclTensor* k = nullptr;
  aclTensor* out = nullptr;
  std::vector<float> logitsHostData = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12};
  std::vector<float> pHostData = {0.2, 0.4, 0.6};
  std::vector<int32_t> kHostData = {1, 2, 3};
  std::vector<float> outHostData(12, 0);
  // Create the logits aclTensor.
  ret = CreateAclTensor(logitsHostData, logitsShape, &logitsDeviceAddr, aclDataType::ACL_FLOAT, &logits);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the p aclTensor.
  ret = CreateAclTensor(pHostData, pShape, &pDeviceAddr, aclDataType::ACL_FLOAT, &p);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the k aclTensor.
  ret = CreateAclTensor(kHostData, kShape, &kDeviceAddr, aclDataType::ACL_INT32, &k);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 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. Replace with the specific API name.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // Call the first-phase API of aclnnApplyTopKTopP.
  ret = aclnnApplyTopKTopPGetWorkspaceSize(logits, p, k, out, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnApplyTopKTopPGetWorkspaceSize 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 aclnnApplyTopKTopP.
  ret = aclnnApplyTopKTopP(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnApplyTopKTopP failed. ERROR: %d\n", ret); return ret);

  // 4. (Boilerplate) Synchronously 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 the device memory to the host memory. Modify based on the specific API definition.
  auto size = GetShapeSize(outShape);
  std::vector<float> 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: %f\n", i, resultData[i]);
  }

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

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

  return 0;
}