aclnnMoeTokenPermute

📄 View Source Code

Applicable Products

Function

• Description: MoE permute computation, which broadcasts and sorts tokens based on the indices.

• Formula:

  • When paddedMode is false, the formula is as follows, where topK refers to the number of experts selected by a token. When the Indices dimension is 2, topK equals the size of the last dimension of Indices; when the Indices dimension is 1, topK equals 1:

    $$sortedIndicesFirst=argSort(\text{flatten}(Indices))$$ $$sortedIndicesOut=argSort(sortedIndicesFirst)$$ $$permuteTokensOut[sortedIndicesOut[i]]=tokens[i//topK]$$

  • When paddedMode is true (not supported yet):

    $$permuteTokensOut[i]=tokens[indices[i]]$$ $$sortedIndicesOut=indices$$

Function Prototype

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

aclnnStatus aclnnMoeTokenPermuteGetWorkspaceSize(
    const aclTensor  *tokens,
    const aclTensor  *indices,
    int64_t           numOutTokens,
    bool              paddedMode,
    const aclTensor  *permuteTokensOut,
    const aclTensor  *sortedIndicesOut,
    uint64_t         *workspaceSize,
    aclOpExecutor   **executor)

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

aclnnMoeTokenPermuteGetWorkspaceSize

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  tokens	Input	Input token features.	Supports empty tensors. Requires a tensor with at least two dimensions, where the size of the first dimension is num_tokens.	FLOAT16, BFLOAT16, FLOAT32	ND	≥2	√
  indices	Input	Input indices index.	Empty tensors are supported. The shape must be 2D or 1D. When paddedMode is false, it indicates the topK expert indices for each input token, with shape (num_tokens, topK) or (num_tokens). When paddedMode is true, it indicates the token indices selected by each expert (not supported yet). The number of elements is less than 16777215, and the values are greater than or equal to 0 and less than 16777215.	INT32, INT64	ND	1 or 2	√
  numOutTokens	Input	Number of valid output tokens.	The value range is any integer. 0 indicates that no tokens are removed. A value greater than 0 slices the tokens sorted by expert based on numOutTokens, retaining the first numOutTokens tokens. A value less than 0 processes based on a negative slice index.	-	-	-	-
  paddedMode	Input.	Indicates whether it is in padding mode.	The value can be false or true. false: non-padding mode, where indices are sorted. true: padding mode, where indices have been padded to represent the token indices selected by each expert, and indices are not sorted in this case (not supported yet).	-	-	-	-
  permuteTokensOut	Output	Tokens expanded and sorted according to indices.	Supports empty tensor. Must be a tensor with at least two dimensions, where the size of the first dimension is min(num_tokens * topK, numOutTokens). The product of the sizes of all dimensions except the first must be the same as the product of the sizes of all dimensions of tokens except the first. The data type is the same as tokens.	FLOAT16, BFLOAT16, FLOAT32	ND	≥2	×
  sortedIndicesOut	Output	Mapping relationship between permuteTokensOut and tokens.	Supports empty tensor. Must be a 1D tensor with shape (num_tokens*topK).	INT32	ND	1	×
  workspaceSize	Output	Returns the workspace size to be allocated 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 Value	Error Code	Description
  ACLNN_ERR_PARAM_NULLPTR	161001	The input and output tensors are null pointers.
  ACLNN_ERR_PARAM_INVALID	161002	The data type of the input or output is not within the supported range.
  ACLNN_ERR_INNER_TILING_ERROR	561002	The shape dimension of tokens is less than 2.
  		The shape of indices is not 1D or 2D, or when paddedMode is false, the first dimension of indices is not equal to the first dimension of tokens.
  		paddedMode is true.

aclnnMoeTokenPermute

• 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 from the first-phase API aclnnMoeTokenPermuteGetWorkspaceSize.
  executor	Input	The op executor, which 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:

  • aclnnMoeTokenPermute 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 "acl/acl.h"
#include "aclnnop/aclnn_moe_token_permute.h"
#include <iostream>
#include <vector>

#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 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);
    // Copy data from the host side to the device side 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);
    // 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];
    }
    // Create an aclTensor by calling the aclCreateTensor API.
    *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: device/stream initialization. See the ACL external API list.
    // Fill in the deviceId based on your actual device.
    int32_t deviceId = 0;
    aclrtStream stream;
    auto ret = Init(deviceId, &stream);
    // Handle the check based on your own needs.
    CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
    // 2. Construct the inputs and outputs based on the API definition.
    std::vector<int64_t> xShape = {3, 4};
    std::vector<int64_t> idxShape = {3, 2};
    std::vector<int64_t> expandedXOutShape = {6, 4};
    std::vector<int64_t> idxOutShape = {6};
    void* xDeviceAddr = nullptr;
    void* indicesDeviceAddr = nullptr;
    void* expandedXOutDeviceAddr = nullptr;
    void* sortedIndicesOutDeviceAddr = nullptr;
    aclTensor* x = nullptr;
    aclTensor* indices = nullptr;
    int64_t numTokenOut = 0;
    bool padMode = false;

    aclTensor* expandedXOut = nullptr;
    aclTensor* sortedIndicesOut = nullptr;
    std::vector<float> xHostData = {0.1, 0.1, 0.1, 0.1, 0.2, 0.2, 0.2, 0.2, 0.3, 0.3, 0.3, 0.3};
    std::vector<int> indicesHostData = {1, 2, 0, 1, 0, 2};
    std::vector<float> expandedXOutHostData = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
    std::vector<int> sortedIndicesOutHostData = {0, 0, 0, 0, 0, 0};
    // Create the self aclTensor.
    ret = CreateAclTensor(xHostData, xShape, &xDeviceAddr, aclDataType::ACL_BF16, &x);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(indicesHostData, idxShape, &indicesDeviceAddr, aclDataType::ACL_INT32, &indices);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // Create the out aclTensor.
    ret = CreateAclTensor(expandedXOutHostData, expandedXOutShape, &expandedXOutDeviceAddr, aclDataType::ACL_BF16, &expandedXOut);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(sortedIndicesOutHostData, idxOutShape, &sortedIndicesOutDeviceAddr, aclDataType::ACL_INT32, &sortedIndicesOut);
    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 aclnnMoeTokenPermute.
    ret = aclnnMoeTokenPermuteGetWorkspaceSize(x, indices, numTokenOut, padMode, expandedXOut, sortedIndicesOut, &workspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMoeTokenPermuteGetWorkspaceSize 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 aclnnMoeTokenPermute.
    ret = aclnnMoeTokenPermute(workspaceAddr, workspaceSize, executor, stream);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMoeTokenPermute failed. ERROR: %d\n", ret); return ret);
    // 4. Boilerplate: synchronize and wait for the task 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 expandedXSize = GetShapeSize(expandedXOutShape);
    std::vector<float> expandedXData(expandedXSize, 0);
    ret = aclrtMemcpy(expandedXData.data(), expandedXData.size() * sizeof(expandedXData[0]), expandedXOutDeviceAddr, expandedXSize * 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 < expandedXSize; i++) {
        LOG_PRINT("expandedXData[%ld] is: %f\n", i, expandedXData[i]);
    }
    auto sortedIndicesSize = GetShapeSize(idxOutShape);
    std::vector<int> sortedIndicesData(sortedIndicesSize, 0);
    ret = aclrtMemcpy(sortedIndicesData.data(), sortedIndicesData.size() * sizeof(sortedIndicesData[0]), sortedIndicesOutDeviceAddr, sortedIndicesSize * sizeof(int32_t),
                      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 < sortedIndicesSize; i++) {
        LOG_PRINT("sortedIndicesData[%ld] is: %d\n", i, sortedIndicesData[i]);
    }
    // 6. Release aclTensor and aclScalar. Modify based on the specific API definition.
    aclDestroyTensor(x);
    aclDestroyTensor(indices);
    aclDestroyTensor(expandedXOut);
    aclDestroyTensor(sortedIndicesOut);

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