aclnnMoeTokenPermuteGrad

📄 View Source Code

Applicable Products

Description

• Function: Performs backward propagation computation for aclnnMoeTokenPermute.

• Formula:

  $$inputGrad = permutedOutputGrad.indexSelect(0, sortedIndices)$$ $$inputGrad = inputGrad.reshape(-1, topK, hiddenSize)$$ $$inputGrad = inputGrad.sum(dim = 1)$$

Function Prototype

Each operator is divided into a two-phase API. You must first call the "aclnnMoeTokenPermuteGradGetWorkspaceSize" API to obtain the input parameters and calculate the required workspace size and the executor that contains the operator computation process based on the computation flow, and then call the "aclnnMoeTokenPermuteGrad" API to perform the computation.

aclnnStatus aclnnMoeTokenPermuteGradGetWorkspaceSize(
    const aclTensor *permutedOutputGrad,
    const aclTensor *sortedIndices,
    int64_t          numTopk,
    bool             paddedMode,
    aclTensor       *out,
    uint64_t        *workspaceSize,
    aclOpExecutor  **executor)

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

aclnnMoeTokenPermuteGradGetWorkspaceSize

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  permutedOutputGrad	Input	Gradient of the forward output permutedTokens.	Requires a 2D tensor, where tokens_num is the number of tokens and topK_num is the value of numTopk.	BFLOAT16, FLOAT16, FLOAT32	ND	[tokens_num * topK_num, hidden_size]	√
  sortedIndices	Input	Sorted index values.	The value range is (0, tokens_num * topK_num - 1), with no duplicate indices.	INT32	ND	tokens_num * topK_num	√
  numTopk	Input	Number of selected experts.	numTopk <= 512.	INT64	ND	-	√
  paddedMode	Input	Whether to enable paddedMode.	true indicates that paddedMode is enabled, and false indicates that paddedMode is disabled. Currently, only false is supported.	bool	-	-	-
  out	output	Gradient of the input tokens.	-	BFLOAT16, FLOAT16, FLOAT32.	ND	(tokens_num, hidden_size)	√
  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 Value	Error Code	Description
  ACLNN_ERR_PARAM_NULLPTR	161001	A required input, required output, or required attribute is a null pointer.
  ACLNN_ERR_PARAM_INVALID	161002	The data type or data format of the input or output is not within the supported range.

aclnnMoeTokenPermuteGrad

• 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 aclnnMoeTokenPermuteGradGetWorkspaceSize.
  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:

  • aclnnMoeTokenPermuteGrad defaults to a deterministic implementation.

• Atlas A2 training products/Atlas A2 inference products, Atlas A3 training products/Atlas A3 inference products: numTopk <= 512.

• Ascend 950PR/Ascend 950DT:

When this API is called, the framework internally calls the aclnnMoeInitRoutingV2Grad API. If a parameter error occurs, refer to the following parameter mapping relationship:

• The permutedOutputGrad input equivalence is the gradExpandedX input of the aclnnMoeInitRoutingV2Grad API.

• The sortedIndices input equivalence is the expandedRowIdx input of the aclnnMoeInitRoutingV2Grad API.

• The numTopk input equivalence is the topK input of the aclnnMoeInitRoutingV2Grad API.

• The paddedMode input equivalence is the dropPadMode input of the aclnnMoeInitRoutingV2Grad API.

  • The out output is equivalent to the out output of the aclnnMoeInitRoutingV2Grad API.

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_grad.h"
#include <iostream>
#include <vector>
#include <cstdio>

#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;
}

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 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);

  // 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 manual.
  // Fill in the 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. Custom construction is required based on the API.

  int64_t num_topk = 2;
  std::vector<float> permuted_output_grad_Data = {1, 2, 3, 4};
  std::vector<int64_t> permuted_output_grad_Shape = {2, 2};
  void *permuted_output_grad_Addr = nullptr;
  aclTensor *permuted_output_grad = nullptr;

  ret = CreateAclTensor(permuted_output_grad_Data, permuted_output_grad_Shape,
                        &permuted_output_grad_Addr, aclDataType::ACL_FLOAT,
                        &permuted_output_grad);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  std::vector<int> sortedIndicesData = {0, 1};
  std::vector<int64_t> sortedIndicesShape = {2};
  void *sortedIndicesAddr = nullptr;
  aclTensor *sortedIndices = nullptr;

  ret = CreateAclTensor(sortedIndicesData, sortedIndicesShape, &sortedIndicesAddr,
                      aclDataType::ACL_INT32, &sortedIndices);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  std::vector<float> outData = {0, 0};
  std::vector<int64_t> outShape = {1, 2};
  void *outAddr = nullptr;
  aclTensor *out = nullptr;

  ret = CreateAclTensor(outData, outShape, &outAddr, 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 aclnnMoeTokenPermuteGrad.
  ret = aclnnMoeTokenPermuteGradGetWorkspaceSize(permuted_output_grad, sortedIndices,
                                                 num_topk, false,
                                                 out, &workspaceSize, &executor);
  CHECK_RET(
      ret == ACL_SUCCESS,
      LOG_PRINT("aclnnMoeTokenPermuteGradGetWorkspaceSize 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 aclnnMoeTokenPermuteGrad.
  ret = aclnnMoeTokenPermuteGrad(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS,
            LOG_PRINT("aclnnMoeTokenPermuteGrad 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 this step based on the specific API definition.
  PrintOutResult(outShape, &outAddr);

  // 6. Release aclTensor and aclScalar. Modify based on the specific API definition.
  aclDestroyTensor(permuted_output_grad);
  aclDestroyTensor(sortedIndices);
  aclDestroyTensor(out);

  // 7. Release device resources.
  aclrtFree(permuted_output_grad_Addr);
  aclrtFree(sortedIndicesAddr);
  aclrtFree(outAddr);

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

  return 0;
}