Data flow analysis for faster boxed arithmetics?

After having written a first `naive' JIT compiler for Smalltalk, I'm
thinking of applying data flow analysis to the bytecodes, in order to
understand when an arithmetic operation cannot cause overflows and hence
will have boxed results format too. As an example of this, `for' loops
(#to:do: in Smalltalk parlance) should be immensely faster.


Did anybody do this (even in other contexts such as Scheme interpreters)
or do you have any pointers to papers about this? After all, boxed
arithmetics have been used for some thirty years.


--

Return to the comp.compilers page.
Search the comp.compilers archives again.