Method
and Apparatus for Distributed Predicate Prediction
Introduction
Predication is a technique to convert control flow to data
flow in modern architectures, but had been used most heavily
in VLIW and dataflow architectures. However, predication is
no panacea. Current technologies have to adopt from one of
the two undesirable options. First, they must avoid predication,
which leaves every control decision as a branch and precludes
distributing instructions over a large, distributed processor.
The second alternative is to predicate instructions, but to
centralize fetch and predicate prediction in a single unit,
resulting in a low fetch bandwidth and, again, the preclusion
of distribution over a large, distributed processor.
Benefits:
- Enables effective predicate prediction in a distributed
uniprocessor environment
- Results in a speedup averaging 20% for only a small additional
amount of state and complexity
Market Potential/Applications:
High-performance, low-power data processing systems
Contact:
University of Texas,
Austin, USA
Website : www.otc.utexas.edu

|