A Feasibility Study on Programmer Specific Instruction Set Processors (PSISPs)

TL;DR

This study evaluates whether programmer-specific instruction set processors (PSISPs) are feasible by analyzing instruction similarity in programmer-specific versus application-specific programs, finding limited significance for programmer-specific instructions.

Contribution

It introduces a method to assess the significance of programmer-specific instructions and evaluates their impact on program similarity measures.

Findings

Programmer-specific instructions have limited impact on program similarity.

Application-specific programs show greater similarity than programmer-specific ones.

The significance of programmer-specific instructions is generally small compared to application-specific instructions.

Abstract

ASIPs are designed in order to execute instructions of a particular domain of applications. The designing of ASIPs addresses the major challenges faced by a system on chip such as size, cost, performance and energy consumption. The higher the number of similar instructions within the domain to be mapped the lesser the energy consumption, the smaller the size and the higher the performance of the ASIP. Thus, designing processors for domains with more similar programs would overcome these issues. This paper describes the investigation of whether the domains of programmer specific programs have any significance like application specific program domains and thus, whether the approach of designing processors known as Programmer Specific Instruction Set Processors is worthwhile. We performed the evaluation at the instruction level by using four different measures to obtain the similarity of programs: (1) by the existence of each instruction, (2) by the frequency of each instruction, (3) by two consecutive instruction patterns and (4) by three consecutive instruction patterns of application specific and programmer specific programs. We found that although programmer specific instructions show some impact on the similarity measures, they are much smaller and therefore insignificant compared to the impact from application specific programs.