Systematic prototyping of superscalar computer architectures

There are a staggering number of design decisions that must be made to design a computer architecture. These are typically resolved using simulation techniques that use inputs composed of a set of traces of instructions taken from executing programs. These traces are collectively called the workload. The workload dictates the architectural-level design. The bottleneck in this architectural design process is the the detailed simulation. One solution that has is to limit the size of the workload used. Although this improves the speed of the design process, it limits the quality of the final design. Another approach is to use published results, yet little or no coherence exists between the workloads that one architecture researcher uses and those used by another. This paper argues that the correct solution to the computer architecture design process is a prototyping approach. In this approach, the initial design is selected using the workload itself. This design is an architectural prototype, specifying the high-level design decisions that are difficult to acquire via detailed simulation. These design decisions include what the mix of function units in the execution stage of the processor is and what the dimensions of the caches in the memory subsystem should be. This paper selects architectural prototypes by trading off accuracy in hardware simulation for an increase in usable workload size.