Interesting candidate predicates for native compilation

[triska]

With great interest, I follow the recent native compilation work of @aarroyoc, available in the cranelift-jit branch.

What are interesting candidate predicates for native compilation, where often a lot of time is spent and which would likely help improve performance measurably in many applications? Personally, I think the following predicates are suitable candidates for trying and evolving this new feature:

1. the predicates of library(assoc), especially get_assoc/3, and insert/5 and the predicates it uses. Since library(assoc) is heavily used by library(clpb) with very large AVL trees, this would likely measurably improve performance on various benchmarks such as the one in Where is the time spent? #207.
2. several rather self-contained and frequently needed predicates in library(clpz), such as
   • cis_gt/2, cis_geq/2, cis_min/3 and other compactified arithmetic operations.
   • building on that, domain_contains/2, domain_remove/3 and other domain_... predicates which are frequently used for reasoning about domains.
   • variable selection predicates for labeling, such as find_ff/4
   • "heavy" parts of propagators, such as relation_unifiable/5 and maximum_matching/1
3. format_cell//1 in library(format).

I hope this is useful. Thank you a lot @aarroyoc for working on this feature, I think this will attract even more attention to Scryer!

[aarroyoc]

This is a good list. Thanks for creating it! I'll keep it as a reminder of the features I need to implement to make them faster.