For memory constrained environments like embedded systems, optimization for program size is often as important, if not more important, as optimization for execution speed. Commonly, compilers try to reduce the code segment and neglect the stack segment, although the stack can significantly grow during the execution of recursive functions as a separate activation record is required for each recursive call. An activation record holds administrative data like the return address and the frame pointer but also the function's formal parameter list and local variables.If a formal parameter or local variable is dead at all recursive calls, then it can be declared globally so that only one instance exists independent of the call depth. We found that in 70% of popular recursive algorithms and in all our real world benchmarks, it is possible to reduce the stack size by declaring formal parameters and local variables globally.Architectures might impose a penalty in code size for accessing global data. On IA32, this stack size reduction starts to materialize for our benchmarks no later than in the fifth recursion.

No abstract

No abstract

No abstract

For memory constrained environments like embedded systems, optimization for program size is often as important, if not more important, as optimization for execution speed. Commonly, compilers try to reduce the code segment size and neglect the stack segment, although the stack can significantly grow during the execution of recursive functions as a separate activation record is required for each recursive call. An activation record holds administrative data like the return address and the frame pointer but also the function's parameter list and local variables.If a formal parameter or local variable is dead at all recursive calls, then it can be declared globally so that only one instance exists independent of the call depth. This allowed us to optimize the stack size in 70% of our benchmarks.Often, live ranges of parameters and local variables can be split at recursive calls through program transformations. Such program transformations allowed us to further optimize the stack size of all our benchmarks.