A physical-level study of the compacted matrix instruction scheduler for dynamically-scheduled superscalar processors

This work studies physical-level characteristics of the recently proposed compacted matrix instruction scheduler for dynamically-scheduled, superscalar processors. Previous work focused on the matrix scheduler´s architecture and argued in support of its speed and scalability advantages. However, no physical-level implementation or models were reported for it. Using full-custom layouts in a commercial 90 nm fabrication technology, this work investigates the latency and energy variations of the compacted matrix and its accompanying logic as a function of the issue width, the window size, and the number of global recovery checkpoints. This work also proposes an energy optimization that throttles unnecessary pre-charges and evaluations. This optimization reduces energy by 10% and 18% depending on the scheduler size.