Phased synthesis of divide and conquer programs

Farzan, Azadeh; Nicolet, Victor

doi:10.1145/3453483.3454089

Cited by 7 publications

(5 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There are two versions of Parsynt available, where different transformation systems are used. We denote them as Parsynt17 [Farzan and Nicolet 2017] and Parsynt21 [Farzan and Nicolet 2021], and consider both of them in evaluation.…”

Section: Baseline Solversmentioning

confidence: 99%

“…D 𝐷 includes all tasks (36 in total) that target to synthesize divideand-conquer-style parallel program from these benchmarks. As a result, D 𝐷 contains all tasks used by Bird [1989]; Farzan and Nicolet [2017] and 12 out of 22 tasks used by Farzan and Nicolet [2021]. The other 10 tasks involve a class of divide-and-conquer different from the tactic proposed by Cole [1995], and thus are out of the scope of AutoLifter.…”

Section: Datasetmentioning

confidence: 99%

“…This will be future work. The only failed task is longest_odd_(0+1) constructed by Farzan and Nicolet [2021], where AutoLifter successfully finds a correct lifting scheme but PolyGen fails.…”

Section: Rq1: Comparison On Divide-and-conquermentioning

confidence: 99%

“…There are two solvers for divide-and-conquer that go beyond the algorithmic tactic used in this paper. Farzan and Nicolet [2019] and Farzan and Nicolet [2021] support programs with nested loops, and the latter also support a more general class of divide operators other than dividing at the The average includes only the results on the tasks solved by both solvers. 2 For results in the form of 𝑎 (𝑏), 𝑎 and 𝑏 represents results under the default setting and the enhanced setting respectively.…”

Section: Case Studymentioning

confidence: 99%

“…On the other hand, implementing a singlepass function is error-prone even for experts. The dataset used by Farzan and Nicolet [2021] contains two bugs that were introduced when the authors manually rewrote the original program into a single-pass program 1 .…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Black-Box Algorithm Synthesis -- Divide-and-Conquer and More

Ji¹,

Xiong²,

Hu³

2022

Preprint

View full text Add to dashboard Cite

Algorithm synthesis is a newly emerging branch of program synthesis, targeting to automatically apply a predefined class of algorithms to a user-provided program. In algorithm synthesis, one popular topic is to synthesize divide-and-conquerstyle parallel programs. Existing approaches on this topic rely on the syntax of the user-provided program and require it to follow a specific format, namely single-pass. In many cases, implementing such a program is still difficult for the user. Therefore, in this paper, we study the black-box synthesis for divide-and-conquer which removes the requirement on the syntax and propose a novel algorithm synthesizer AutoLifter. Besides, we show that AutoLifter can be generalized to other algorithms beyond divide-and-conquer. We propose a novel type of synthesis tasks, namely lifting problems, and show that AutoLifter can be applied to those algorithms where the synthesis task is an instance of lifting problems. To our knowledge, AutoLifter is the first algorithm synthesizer that generalizes across algorithm types. We evaluate AutoLifter on two datasets containing 57 tasks covering five different algorithms. The results demonstrate the effectiveness of Au-toLifter for solving lifting problems and show that though AutoLifter does not access the syntax of the user-provided program, it still achieves competitive performance compared with white-box approaches for divide-and-conquer.

show abstract

Section: Baseline Solversmentioning

confidence: 99%

Section: Datasetmentioning

confidence: 99%

Section: Rq1: Comparison On Divide-and-conquermentioning

confidence: 99%

Section: Case Studymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Black-Box Algorithm Synthesis -- Divide-and-Conquer and More

Ji¹,

Xiong²,

Hu³

2022

Preprint

View full text Add to dashboard Cite

show abstract

FlashFill++: Scaling Programming by Example by Cutting to the Chase

Cambronero

Gulwani

et al. 2023

Proc. ACM Program. Lang.

View full text Add to dashboard Cite

Programming-by-Examples (PBE) involves synthesizing an "intended program" from a small set of user-provided input-output examples. A key PBE strategy has been to restrict the search to a carefully designed small domain-specific language (DSL) with "effectively-invertible" (EI) operators at the top and "effectively-enumerable" (EE) operators at the bottom. This facilitates an effective combination of top-down synthesis strategy (which backpropagates outputs over various paths in the DSL using inverse functions) with a bottom-up synthesis strategy (which propagates inputs over various paths in the DSL). We address the problem of scaling synthesis to large DSLs with several non-EI/EE operators. This is motivated by the need to support a richer class of transformations and the need for readable code generation. We propose a novel solution strategy that relies on propagating fewer values and over fewer paths. Our first key idea is that of "cut functions" that prune the set of values being propagated by using knowledge of the sub-DSL on the other side. Cuts can be designed to preserve completeness of synthesis; however, DSL designers may use incomplete cuts to have finer control over the kind of programs synthesized. In either case, cuts make search feasible for non-EI/EE operators and efficient for deep DSLs. Our second key idea is that of "guarded DSLs" that allow a precedence on DSL operators, which dynamically controls exploration of various paths in the DSL. This makes search efficient over grammars with large fanouts without losing recall. It also makes ranking simpler yet more effective in learning an intended program from very few examples. Both cuts and precedence provide a mechanism to the DSL designer to restrict search to a reasonable, and possibly incomplete, space of programs. Using cuts and gDSLs, we have built FlashFill++, an industrial-strength PBE engine for performing rich string transformations, including datetime and number manipulations. The FlashFill++ gDSL is designed to enable readable code generation in different target languages including Excel's formula language, PowerFx, and Python. We show FlashFill++ is more expressive, more performant, and generates better quality code than comparable existing PBE systems. FlashFill++ is being deployed in several mass-market products ranging from spreadsheet software to notebooks and business intelligence applications, each with millions of users.

show abstract