Out-of-order Predicated Execution with Translation Register Buffer
Abstract
New generation superscalar processors combine predication with large resources. A typical example is the EPIC architecture as defined by the canonical IA64 ISA. Unlike traditional ISAs, these new instruction sets are resistant to an out-of-order execution engine, because of the resource size as well as the complexity of executing predicated instructions. In this paper, we present a novel register management policy that facilitates the out-of-order execution of a fully predicated ISA. For this purpose, a new mechanism, called Translation Register Buffer (TRB) is introduced. A translation register acts as an intermediate register that associates a logical register with a physical register. By providing an indirect access to a logical register, the translation register supports the side effects induced by the cancellation of instructions by predicates. In order to demonstrate the TRB validity, a complete simulation framework that fully supports the IA64 ISA has been designed. This original implementation features an emulator and a cycle accurate 10-stage out-of-order core simulator. Our simulation results indicate that, on average, a 10 be achieved, as compared with the equivalent in-order EPIC architecture.
Domains
Other [cs.OH]
Loading...