Resumen de Content aware architectures
This thesis presents a dissertation in Content Aware Architectures, They exploit knowledge on data content from architectural point of view.
Previous studies about data content were successful from several different points of view, but possibly the most important studies were based on the locality principle.
In the decades since introduction of the locality principle several important computer architectural improvements were achieved, but most of them were from memory hierarchy point of view: caches and TLB are some of main examples.
This thesis defines architectural improvements derived from this principle in other parts of the processor. They are based on a new data taxonomy that grew up from a concept called d- similarity. D-similarity can be seen as another kind of locality value locality.
New base mechanism was designed to take benefit from d-similarity on integer values: Content Aware Base Mechanism (CABase Mechanism). It will be main mechanism for rest of proposals of this Thesis.
First natural extension of processor architecture taking advantage of d-similarity was in register file. This Register File was called Content Aware Register File (CARF). CARF reduces power consumption with minimal impact on performance and is orthogonal to other microarchitectural mechanisms used to reduce power in register files.
CARF results about data type of source operands (of instructions) showed that execution flow could be divided between two flows: clustered architecture. This architecture is an asymmetric clustered architecture, where every cluster is adapted following d-similarity taxonomy. The new architecture is called Content Aware Asymmetric Cluster (CAAC).
Other extension that required minimal changes after the introduction of CARF was in Translation Lookaside Buffer (TLB). This extension, called Content Aware TLB (CATLB), provided a mechanism to satisfy many of TLB accesses via what we called a 0-cycle TLB. While Content Aware Register File demonstrated that this new point of view on locality could be applied to other parts of processor architecture, Content Aware TLB showed that it was also possible to obtain benefits in the memory hierarchy.
This thesis has shown that this new point of view on the locality principle can get further performance and reduce complexity.These group of architectures are called Content Aware Architectures.
