Hardware to Software Migration with Real-Time Thread Integration

This paper introduces thread integration, a new method of providing low-cost concurrency for microcontrollers and microprocessors. This post-pass compiler technology effectively interleaves multiple threads of control at the instruction level for execution on a uniprocessor and implicitly provides very fine-grain multithreading without context switching overhead. Such efficient concurrency allows implementation of real-time functions in software rather than dedicated peripheral hardware. Thread integration's position in the design space is evaluated to help identify when and where to apply the technique. A number of code transformations have been developed that allow a guest thread of computation to be systematically integrated into a host thread while meeting all the real-time requirements of the guest thread. The thread integration concept and the associated code transformations have been successfully applied to several example applications. Designers of embedded systems must often deal with very restrictive cost, power, size and weight constraints. Recurring costs become very significant as production volumes reach millions of units per year, encouraging manufacturers to invest in substantial up-front engineering to reduce costs. Battery operated equipment is very sensitive to power consumption as battery-life expectations grow in the marketplace. Size and weight constraints affect portable electronics as well as subsystems for avionics, automotive and other applications. Real-time requirements in system specifications can exacerbate these constraints by forcing the use of additional hardware or software dedicated to meeting real-time task deadlines. Dedicated peripheral hardware can be added to the system to perform the time-critical operations. However, this addition increases system cost, makes the system larger and heavier and increases power consumption. The time critical operations can be implemented in software, given a sufficiently fast processor. Processing a single task with events scheduled through busy waiting becomes inefficient as idle times rise when event frequencies fall. Task switching allows processing resources to be shared with another task, but the overhead incurred by saving state information reduces efficiency and limits the maximum switching rate. Integrated circuit fabrication technology and architectural advances continue to reduce the cost of microprocessor compute cycles. As this trend continues, the recurring cost penalty of implementing functions in dedicated peripheral hardware rather than in software on the microprocessor will rise, encouraging software implementations. This paper presents the concept of thread integration, a compile-time technique for interleaving multiple threads of computation at the instruction level and hence providing multithreading on a generic uniprocessor without context switching overhead. This technique uses the ever-cheaper compute …