Documentation

Development Workflow and Key APIs

Scope fusion scenarios are classified into two scenarios: many-to-one and many-to-many. This topic describes the development workflow and key APIs in the two scenarios.

Many-to-One Scope Fusion

In the many-to-one scenario, multiple small operators in the scope are fused into a large operator.

Figure 1 Implementation workflow in the many-to-one scenario

Many-to-Many Scope Fusion

In the many-to-many scenario, multiple small operators in the scope are fused into a composite of small operators.

Figure 2 Implementation workflow in the many-to-many scenario

The many-to-one and many-to-many fusion scenarios are basically the same. The differences are described as follows:

  1. In the fusion result setting function GenerateFusionResult, you need to set the connections between internal operator composites.
  2. The internal small operators are not parsed by the plugin. Therefore, you need to construct IRs for the small operators and set required parameters such as attributes.

The following uses the many-to-one scope fusion scenario as an example to describe how to design fusion patterns and develop fusion specifications.