ENABLE_ACLNN

Description

This environment variable will be deprecated in later versions.

During graph build, this environment variable can be set to determine whether to call the host execution function registered by the operator during graph execution to implement the host execution logic and kernel delivery. The values are as follows:

  • true: During graph execution, the host execution function registered by the operator is called to implement the host execution logic and kernel delivery.
  • false: The GE executor is still used to deliver the kernel of the operator during graph execution.

The details are as follows:

  • If the operator supports only the API calling by a single-operator on the host (aclnnSupport in the information library is set to aclnn_only), this environment variable does not take effect, and the host execution function is always called to implement the host execution logic and kernel delivery.
  • If the operator does not support the API calling by a single-operator on the host (the aclnnSupport field is not configured in the information library), the GE executor is still used to deliver the kernel of the operator.
  • If both modes are supported, the host execution function is called to implement the host execution logic and kernel delivery only when the following three conditions are met: the environment variable value is true, aclnnSupport is configured in the information library (aclnnSupport is set to support_aclnn), and dynamic shape (the shape field contains a -1 or -2 placeholder) is used. Otherwise, the GE executor is still used to deliver the kernel of the operator.

Example

export ENABLE_ACLNN=true

Restrictions

None

Applicability

Atlas training products

Atlas inference products

Atlas A2 training products/Atlas A2 inference products

Atlas A3 training products/Atlas A3 inference products