[object Object][object Object][object Object]undefined[object Object]

Synchronizes all ranks in a communication domain. xRef is used only to build tensor dependencies. No operation is performed on xRef in the API.

Compared with the [object Object] API, this API has the following changes:

• Dynamic scale-in support: The operator can run normally without recompilation after faulty ranks are removed after the communication domain is created. Enable this feature by passing the [object Object] parameter.
• Timeout detection support:
  • Enable this feature by passing a [object Object] parameter greater than 0 microseconds, with a maximum supported value of INT32_MAX. When the internal synchronization wait time of the operator exceeds the given [object Object], a timeout exception occurs on the rank that runs the operator.

[object Object]

Each operator has calls. First, [object Object] is called to obtain the workspace size required for computation and the executor that contains the operator computation process. Then, [object Object] is called to perform computation.

[object Object]

• Parameters

  [object Object]
  • [object Object]Atlas A2 training products/Atlas A2 inference products[object Object]: For these products, [object Object] and [object Object] only takes null pointers.

• Returns

  aclnnStatus status code. For details, see .

  The first-phase API implements input parameter verification. The following errors may be thrown.

  [object Object]

[object Object]

• Parameters

  [object Object]

• Returns

  aclnnStatus status code. For details, see .

[object Object]

• Deterministic computing:

  • [object Object] defaults to a deterministic implementation.

• Constraints on the use of the communication domain:

  • [object Object] in a model requires an independent communication domain, and no other operators are allowed in the communication domain.

• Application scenarios:

  • Calling this operator in a network model that requires all-rank synchronization can mask performance fluctuations caused by fast and slow ranks, assisting in performance analysis.
  • This operator can be called continuously. When building a graph, the input of the previous operator and the output of the next operator should be passed as input parameters to the API.
  • The dynamic scale-in feature cannot be enabled in the tensor parallelism scenario.

• Parameter consistency constraint:

  • When [object Object] is enabled, ensure that [object Object] and [object Object] or [object Object] also enable this parameter, and their values remain consistent with the corresponding [object Object] parameter.

[object Object]

• Prepare files.

  1. Create a [object Object] directory. Follow the instructions to create [object Object] and [object Object] files in the [object Object] directory and modify them according to the code.

  2. Install the CANN package and compile and run barrierDemo.

• Compilation script:

  [object Object]

• Compilation and execution:

  [object Object]

• The example code is as follows:

  [object Object]