Advanced compiling techniques to reduce RAM usage of static operating systems

In recent years, a rapidly growing number of small embedded systems have been used in very high volumes. One example is the automotive industry, where the number of Electronic Control Units (ECU) in a single car is approaching 100 for high end automobiles and several dozens are used in mid-range cars. Small system-on-chip microcontrollers are often used with static operating systems. As on-chip RAM is rather expensive and only few KBs of RAM are available on such devices, reducing the RAM usage is an important objective in order to save costs — especially in high-volume production. This thesis presents several new approaches to reduce the RAM usage of such systems by applying advanced compilation and optimization techniques. Common optimizations are examined regarding their impact on RAM usage. By selecting classical optimization algorithms regarding their impact on RAM usage, the RAM required for a series of test cases is reduced by almost 20%. Upper bounds for stack sizes of application tasks will be statically calculated using high-level analysis available in the compiler. Comparisons with a commercial tool working on machine-code-level show clear advantages regarding maintainability as well as reliability. Most important, the register sets stored by the operating system when a task is preempted are optimized by abstaining from saving unnecessary registers. Inter-task register-allocation further reduces the RAM required to save those task contexts. The new algorithms have been added to a production quality compiler and a full commercial OSEK implementation was modified to make use of the new optimizations. Tests on real hardware as well as comparisons with commercial tools not only show that the system works and improves usability and maintainability, but also that significant reductions of RAM requirements, and therefore cost savings, are possible. In a series of benchmarks, RAM usage is reduced on average by 30%–60%.