4. Multi-pipeline model

Order-based superscalars have an execution model shown in figure 8 . Pipelines break down into two parts:

• a first part, sometimes called "static", acquires several instructions (sequential or from the branch target), decodes them, examines whether hardware resources are available, detects any data dependencies, and places in a launch buffer those instructions that can be launched in the same clock cycle. The maximum number of instructions that can be launched is usually 2 or 4. A number of rules define which instructions can be started in the same cycle. Stages in this first part of the pipeline may be suspended when resources are not available, when data dependencies exist, or when certain complex instructions need to be broken down into several...