aclnnMoeTokenPermuteWithEp

📄 View Source Code

Applicable Products

Function

• Description: The MoE permute operation. It broadcasts tokens and optional probs based on indices, sorts them, and slices them according to the range specified in rangeOptional.

• Formula:

  • When paddedMode is false, the formula is as follows. topK indicates the number of experts selected for each token. If Indices is 2-dimensional, topK equals the size of the last dimension of Indices. If Indices is 1-dimensional, topK is 1.

    $$sortedIndicesFirst=argSort(\text{flatten}(Indices))$$ $$sortedIndicesOut=argSort(sortedIndicesFirst)$$

    When rangeOptional[0] <= sortedIndicesOut[i] < rangeOptional[1]:

    $$permuteTokensOut[sortedIndicesOut[i] - rangeOptional[1]]=tokens[i//topK]$$

  • When paddedMode is true (not supported yet):

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

Function Prototype

Each operator uses a two-phase API. You must first call the "aclnnMoeTokenPermuteWithEpGetWorkspaceSize" API to obtain the workspace size required for computation and the executor that encapsulates the operator's computation flow, and then call the "aclnnMoeTokenPermuteWithEp" API to perform the computation.

aclnnStatus aclnnMoeTokenPermuteWithEpGetWorkspaceSize(
    const aclTensor    *tokens,
    const aclTensor    *indices,
    const aclTensor    *probsOptional,
    const aclIntArray  *rangeOptional,
    int64_t             numOutTokens,
    bool                paddedMode,
    const aclTensor    *permuteTokensOut,
    const aclTensor    *sortedIndicesOut,
    const aclTensor    *permuteProbsOut,
    uint64_t           *workspaceSize,
    aclOpExecutor     **executor)

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

aclnnMoeTokenPermuteWithEpGetWorkspaceSize

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  tokens	Input	Represents the input tokens in the permute operation.	Supports null tensor. Must be a 2D tensor with the shape (num_tokens, hidden_size).	BFLOAT16, FLOAT16, FLOAT32	ND	2	√
  indices	Input	Indicates the expert indices corresponding to the input tokens.	Supports null tensor. The shape must be 1D or 2D. When paddedMode is false, it indicates the topK expert indices processed for each input token, with shape (num_tokens, topK_num) 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	√
  probsOptional	Input	Represents the expert probabilities corresponding to the input tokens.	Supports null tensor. Corresponds to the permuteProbsOut output. If null is passed, permuteProbsOut is not output. The shape is the same as that of indices.	BFLOAT16, FLOAT16, FLOAT32	ND	1 or 2	√
  rangeOptional	Input	Valid range of the EP split.	Supports null. The size is 2. When null, probsOptional and permuteTokensOut are ignored, and the execution logic falls back to aclnnMoeTokenPermute.	-	-	-	-
  numOutTokens	Input	Number of valid output tokens.	The value range is any integer. 0 indicates that no tokens are removed. When greater than 0, the tokens sorted by expert are sliced according to numOutTokens, retaining the first numOutTokens tokens. When less than 0, processing is performed based on the negative slice index.	-	-	-	-
  paddedMode	Input	Indicates whether it is in padded mode.	The value can be false or true. false: indicates non-padded mode, where indices are sorted. true: indicates padded 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 based on indices.	Supports null tensor. Must be a 2D tensor with shape (rangeOptional[1] - rangeOptional[0], hidden_size). The data type is the same as tokens.	BFLOAT16, FLOAT16, FLOAT32	ND	2	×
  sortedIndicesOut	Output	Mapping between permuteTokensOut and tokens.	Supports null tensor. Must be a 1D tensor with shape (num_tokens * topK_num).	INT32	ND	1	×
  permuteProbsOut	Output	Probs expanded and sorted based on indices.	Supports null tensor. Must be a 1D tensor with shape (rangeOptional[1] - rangeOptional[0]). The data type is the same as tokens.	BFLOAT16, FLOAT16, FLOAT32	ND	1	×
  workspaceSize	Output	Returns the workspace size to be allocated 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	Return Code	Description
  ACLNN_ERR_PARAM_NULLPTR	161001	The input or output tensor is a null pointer.
  ACLNN_ERR_PARAM_INVALID	161002	The data type or data format of the input and output is not within the supported range.
  ACLNN_ERR_INNER_TILING_ERROR	561002	The shape dimension of tokens is not 2.
  		The shape of indices is not 1D or 2D, or when paddedMode is false, the first dimension of the shape of indices is not equal to the first dimension of tokens.
  		paddedMode is true (not supported yet).

aclnnMoeTokenPermuteWithEp

• Parameters:

  Parameter	Input/Output	Description
  workspace	Input	Specifies the workspace memory address applied on the device side.
  workspaceSize	Input	Specifies the workspace size applied on the device side, obtained from the first-phase API aclnnMoeTokenPermuteWithEpGetWorkspaceSize.
  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:

  • aclnnMoeTokenPermuteWithEp 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_with_ep.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 CreateAclIntArray(const std::vector<T>& hostData, void** deviceAddr, aclIntArray** intArray) {
  auto size = GetShapeSize(hostData) * sizeof(T);
  // Call aclrtMalloc to allocate memory on the device.
  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 the data on the host to the memory on the device.
  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);

  // Call aclCreateIntArray to create an aclIntArray.
  *intArray = aclCreateIntArray(hostData.data(), hostData.size());
  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);
    // Copy data from the host to the device 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);
    // 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];
    }
    // 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: initialize the device/stream. See the ACL external API list.
    // Set 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> probsShape = {3, 2};
    std::vector<int64_t> expandedXOutShape = {4, 4};
    std::vector<int64_t> idxOutShape = {6};
    std::vector<int64_t> expandedProbsOutShape = {4};

    void* xDeviceAddr = nullptr;
    void* indicesDeviceAddr = nullptr;
    void* probsDeviceAddr = nullptr;
    void* expandedXOutDeviceAddr = nullptr;
    void* sortedIndicesOutDeviceAddr = nullptr;
    void* expandedProbsOutDeviceAddr = nullptr;
    void* rangeDeviceAddr = nullptr;
    aclTensor* x = nullptr;
    aclTensor* indices = nullptr;
    aclTensor* probs = nullptr;
    aclIntArray* range = nullptr;
    int64_t numTokenOut = 6;
    bool padMode = false;

    aclTensor* expandedXOut = nullptr;
    aclTensor* sortedIndicesOut = nullptr;
    aclTensor* expandedProbsOut = 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, 3, 1, 2, 3};
    std::vector<float> probsHostData = {0.5, 0.3, 0.4, 0.2, 0.5, 0.4};
    std::vector<float> expandedXOutHostData = {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};
    std::vector<float> expandedProbsOutHostData = {0, 0, 0, 0};
    std::vector<int64_t> rangeHostData = {1, 5};
    // 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);
    ret = CreateAclTensor(probsHostData, probsShape, &probsDeviceAddr, aclDataType::ACL_BF16, &probs);
    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);
    ret = CreateAclTensor(expandedProbsOutHostData, expandedProbsOutShape, &expandedProbsOutDeviceAddr, aclDataType::ACL_BF16, &expandedProbsOut);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // Create the related attr.
    ret = CreateAclIntArray(rangeHostData, &rangeDeviceAddr, &range);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // 3. Call the CANN operator library API. Replace this with the specific API.
    uint64_t workspaceSize = 0;
    aclOpExecutor* executor;
    // Call the first-phase API of aclnnMoeTokenPermuteWithEp.
    ret = aclnnMoeTokenPermuteWithEpGetWorkspaceSize(x, indices, probs, range, numTokenOut, padMode, expandedXOut, sortedIndicesOut, expandedProbsOut, &workspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMoeTokenPermuteWithEpGetWorkspaceSize 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 aclnnMoeTokenPermuteWithEp.
    ret = aclnnMoeTokenPermuteWithEp(workspaceAddr, workspaceSize, executor, stream);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMoeTokenPermuteWithEp 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-side memory to the host side. 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]);
    }
    auto expandedProbsSize = GetShapeSize(expandedProbsOutShape);
    std::vector<float> expandedProbsData(expandedProbsSize, 0);
    ret = aclrtMemcpy(expandedProbsData.data(), expandedProbsData.size() * sizeof(expandedProbsData[0]), expandedProbsOutDeviceAddr, expandedProbsSize * 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 < expandedProbsSize; i++) {
        LOG_PRINT("expandedProbsData[%ld] is: %f\n", i, expandedProbsData[i]);
    }
    // 6. Release aclTensor and aclScalar. Modify based on the specific API definition.
    aclDestroyTensor(x);
    aclDestroyTensor(indices);
    aclDestroyTensor(probs);
    aclDestroyTensor(expandedXOut);
    aclDestroyTensor(sortedIndicesOut);
    aclDestroyTensor(expandedProbsOut);

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