Abstractions and small languages in synthesis CS294 : Program Synthesis for Everyone

Publish in



Please download to get full document.

View again

of 36
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Abstractions and small languages in synthesis CS294 : Program Synthesis for Everyone. Ras Bodik Emina Torlak. Division of Computer Science University of California, Berkeley. Today.
Abstractions and small languages in synthesisCS294: Program Synthesis for Everyone Ras Bodik Emina Torlak Division of Computer Science University of California, Berkeley Today Today: we describe why high-level or domain-specific programming abstractions, provided as language constructs, make synthesis more efficient and easier to use. Next lecture: Student presentations (problem stmt). Subsequent lecture: Language implementation Part I. Racket macros. Language embedding. Instructions for classroom presentation Topic: problem statement (refinement of HW1) Elaborate on synthesis artifacts:
  • what will be synthesized
  • what are the specs (this item is important!) 3-minutes per student/team ==> practice! Email Ras.ppt(x) slides by 9am before lecture.
  • Outline Review of HW2 description of staff solution Lessons from HW2 motivate synthesis at high level of abstraction Reducing the candidate space (tree rotation) prune with domain constrains Reducing the formula size (graph classifiers) synthesis followed by code generation Synthesis at functional level (time permitting) followed by data structure generation Advanced challenge Is this lecture familiar material? Entertain yourself by designing a small language L that can
  • express distributed protocols and
  • can model message loss and reordering How to translate programs in L to formulas, or otherwise find L programs that meet a spec. Oh yes, when you are done, what is a good spec language for distributed protocols?
  • HW2 feedback Description of solutions We sped up the encoding by
  • using smallest bit vectors possible for each variable
  • not relying on the extensional theory of arrays
  • eliminating redundant constraints on 2-bit variables represented as bit vectors of length 2
  • eliminating constant arrays
  • replacing macros with explicit let statements; and
  • telling the solver which logic the encoding is in.
  • Lessons (encoding) why using bitvectors helps
  • bounded by the type ==> can save some explicit constraints on values of bitvector variables
  • different decision procedure (eg blasting to SAT) why must also drop Ints?
  • absence of Ints allows bitblasting because no need to reason about (infinite) ints
  • essentially, a different algorithm is used why not relying on extensional theory helps
  • (= a b) insists that entire arrays a,b are equal, which could be infinitely many if indexes are Ints
  • a[0]=b[0] … insists only on bounded number of equalities ==> enumerate what needs to hold
  • Lessons (the input constraint for ind. synth.) one perfect input vs. identify sufficient inputs
  • Def: perfect ==> correct on a perfect input implies correct on all inputs
  • a good input accelerates solving careful about selecting the perfect input
  • we were wrong in Part 2
  • Q: how to overcome the danger of weak input?
  • Results (z3) Results (Kodkod) Wish list from HW2 Wish list:
  • start the solver earlier
  • start the homework earlier
  • use faster solvers
  • get feedback on where the solver is wasting time
  • debug encoding on 2x2 matrix, then scale up
  • facilitate easier tweaking of constraints
  • Reducing the Size of Candidate Space Example: Synthesis of tree rotation We want to suitably rotate tree A into tree B. We don’t know exactly how to rotate. So we ask the synthesizer. p p a b b a c c Partial program for rotation We have to update (up to) 7 memory locations. We have seven pointer values available. A straightforward partial program: r.left := {| p | a | b | c | | | | |} a.left:= {| p | a | b | c | | | | |} … c.right:= {| p | a | b | c | | | | |} Search space: , about 1017 Reducing the search space Encode that the pointer rotation is a permutation. (p.left, a.left, …, c.right) := synth_permutation(p, a, b, c, ,, , ) Search space: Implementing the permutation construct defsynth_permutation(lst): retval = empty list chosen = empty set repeat len(lst) times ix = ??(0..len(lst)-1) append lst[ix] to retval assert ix not in chosen add ix to chosen return retval How many choices exist for len(lst) = 7? so does using the permutation reduce search space to 7! ? Locally ruled out choices In synth_permutation, selecting ix that has been chosen is immediately ruled out by the assertion We call this locally ruled out choice. there are 7!, not 77 ,choices that satisfy the assertion Compare this with a globally ruled out choice such a choice fails only after the solver propagates its effects to assertions in the postcondition. Further space reduction In addition to a permutation, we insist that the reordered nodes form a binary search tree (p.left, a.left, …, c.right) := synth_permutation(p, a, b, c, , , , ) assert bst_to_depth_4(p) def bst_to_depth_4(p): assert p.d >= p.left.d … and p.d <= p.right.right.right.d How is this a small language? What do permutation, bst_to_depth_4 have to do with abstractions or languages? These are constructs of a tree manipulation language We defined them inside the host language ie, they are embedded in the host and compiled to formulas Summary Effective size of candidate space Because local assertions prune the search space In fact, recall L4: more bits in encoding often better Reducing the Size of Encoding Graph classifiers Synthesize graph classifiers (ie, repOK checkers), eg:
  • singly linked list
  • cyclic linked list
  • doubly linked list
  • directed tree
  • tree with parent pointer ---->
  • strongly connected Ensure linear running time. [Izthakyet al, OOPSLA 2010]
  • Specification (tree with parent pointer) Precondition (integrity assumption): root r via C functional R Postcondition (classification): Synthesized linear-time classifier The classifier (not a simple paraphrase of the spec!): Explained: This classifier still looks declarative to me! This classifier can be compiled to an operational pgm. with guaranteed linear time performance First, using DFS, compute inverse edges so that we can compute predecessor sets Next, compute these conditions with DFS: The partial program Recall that a partial program (sketch) is a grammar. each classifier is a <stmt> from this grammar How is linear time guaranteed? The partial program contains only one variable, v hence we cannot form properties over, say, pairs of nodes Reachability across label strings only from the root is legal but is not why? evaluating, say, needs time Regular expressions are bounded in length, of course hence they can be computed during DFS Discussion What did we gain with this high-level program? encoding: solver efficiency: engineering complexity: Their inductive synthesis algorithm Simple thanks to the structure of the language:
  • assume you have positive and negative instance sets P, N.
  • enumerate all clauses C
  • find clauses CP that are true on each graph in P
  • find smallest subset of CP such that is false for all graphs from N
  • Summary of Izhakyet al The key concept we have seen is synthesis at high-level of abstraction
  • guarantees resource constraints (here, linear time)
  • a simpler synthesis algorithm followed by deterministic compilation
  • essentially, this is just pattern-driven code generation
  • eg, translate #pC(v) to some fixed code
  • Other uses of languages? Summary synthesis followed by deterministic compile the compiler could benefit from synthesis, though higher-level abstraction ==> smaller programs and thus smaller formulas not by itself smaller search spaces reduce search space via domain constraints eg, what rotations are legal Concepts not covered constructs for specs, including examples ex: angelic programming could create examples inputs reduce ambiguity if your spec is incomplete (eg examples), then smaller candidate space reduces ambiguity in the spec feedback to the user/programmer in familiar domain eg describing the localized bug using unsat core support abstraction that will be used in synthesis ignore actual value in AG, actual multiplication in HPC codes implicitly codify domain properties
  • so that you can automatically determine that a single matrix is sufficient for xpose
  • Looking ahead Languages that will be built in cs294 projects:
  • distributed protocols (asynchrony, lost messages)
  • distributed protocols (bounded asynchrony)
  • web scraping (how to name DOM elements)
  • spatial programming in forth
  • attribute grammar evaluators
  • distributed memory data structures and operations
  • parsers for programming contests
  • Next lecture (Tuesday) Read Fudging up Racket Implementing a language in Racket Optimizations
    Related Search
    Related Documents
    View more...
    We Need Your Support
    Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

    Thanks to everyone for your continued support.

    No, Thanks