Thread Integration for Error Detection and Performance

This paper presents a technique for integrating multiple threads of computation for simultaneous execution and which is well suited for fault-tolerant application programs. A post-pass compiler has been developed that is capable of taking the application program as the host thread and automatically integrating the additional code as a guest thread to produce a composite thread, the execution of which provides integrated, transparent self-monitoring capability for embedded control systems. This thread integration method is general and can be used to efficiently merge any two execution threads. This allows the composite thread to be executed in a uniprocessor without requiring a multithreaded processor and run-time environment. This also allows the exploitation of idling or underutilized machine resources in modern superscalar processors, designed to exploit instruction-level parallelism within a single thread of execution, for dependability enhancing objectives. Consequently, fault monitoring mechanisms can be automatically incorporated without requiring specialized hardware and without incurring significant performance penalty. Experimental results using the image processing functions from the SDIO Signal and Data Processing Benchmark Suite are used to demonstrate the compilation tool and the thread integration method.