mrsc
mrsc copied to clipboard
ilya-klyuchnikov
A toolkit for building multi-result supercompilers
May be a bit non-trivial to fit the whole MRSC framework, but anyway, is quite interesting.
There are variants of traditional supercompilation (including Turchin) when if current configuration is an instance (but not renaming) of some configuration, then folding is performed. Some people say that in...
In classical supercompilation when whistle blows, generalization (rebuilding) is performed. We can introduce two-level whistle in multi-result supercompilation: something like Warning/Hint. The logic is as follows: we cannot continue driving...
It should allow to build a really huge numbers of graphs.
This is a true research. There are some techniques for graph normalization in literature on labeled transition systems. Why not to apply this technologies?
The idea: to minimize the size of residual program.
See for reference: - Introduction to Combinators and λ-Calculus by Hindley and Seldin
It should be extremely interesting to transform functional program with recursion and pattern matching into lambda calculus, supercompile it and then translate it back. See two papers: - Efficient Interpretation...
It is extremely easy to make a generic component for residuation if residual program is in form: `lhs -> rhs` with scoping. It also should be true top-down residuation: code...
Metadata
Owner
Metadata
A toolkit for building multi-result supercompilers