The presence of equivalent mutants has long been considered a major obstacle to the widespread adoption of mutation analysis and mutation testing. This paper presents a study on the types and prevalence of equivalent mutants in real-world Java programs. We conducted a ground-truth analysis of 1,992 mutants, sampled from 7 open source Java projects. Our analysis identified 215 equivalent mutants, which we grouped based on two criteria that describe why the mutants are equivalent and how challenging their detection is. From this analysis, we observed that (1) the median equivalent mutant rate across the 7 projects is 2.97%; (2) many equivalent mutants are caused by common programming patterns and their detection is not much more complex than structural pattern matching over an abstract syntax tree. Based on the findings of our ground-truth analysis, we developed Equivalent Mutant Suppression (EMS), a technique that comprises 10 efficient and targeted analyses. We evaluated EMS on 19 open- source Java projects, comparing the effectiveness and efficiency of EMS to two variants of Trivial Compiler Equivalence (TCE), the current state of the art in equivalent mutant detection. Additionally, we analyzed all 9,047 equivalent mutants reported by any tool to better understand the types and frequencies of equivalent mutants found. Overall, EMS detects 8,776 equivalent mutants within 325 seconds; TCE detects 2,124 equivalent mutants in 2,938 hours.

Mutation testing offers concrete test goals (mutants) and a rigorous test efficacy criterion, but it is expensive due to vast numbers of mutants, many of which are neither useful nor actionable. Prior work has focused on selecting representative and sufficient mutant subsets, measuring whether a test set that is mutation-adequate for the subset is equally adequate for the entire set. However, no known industrial application of mutation testing uses or even computes mutation adequacy, instead focusing on iteratively presenting very few mutants as concrete test goals for developers to write tests.

Search-based techniques have been demonstrated effective in solving complex optimization problems that arise in domain-specific compilers for machine learning (ML). Unfortunately, deploying such techniques in production compilers is impeded by two limitations. First, prior works require factorization of a computation graph into smaller subgraphs over which search is applied. This decomposition is not only non-trivial but also significantly limits the scope of optimization. Second, prior works require search to be applied in a single stage in the compilation flow, which does not fit with the multi-stage layered architecture of most production ML compilers. This paper presents Xtat, an autotuner for production ML compilers that can tune both graph-level and subgraph-level optimizations across multiple compilation stages. Xtat applies Xtat-M, a flexible search methodology that defines a search formulation for joint optimizations by accurately modeling the interactions between different compiler passes. Xtat tunes tensor layouts, operator fusion decisions, tile sizes, and code generation parameters in XLA, a production ML compiler, using various search strategies. In an evaluation across 150 ML training and inference models on Tensor Processing Units (TPUs) at Google, Xtat offers up to 2.4x and an average 5% execution time speedup over the heavily-optimized XLA compiler.

Utilizing memory and register bandwidth in modern architectures may require swizzles --- non-trivial mappings of data and computations onto hardware resources --- such as shuffles. We develop Swizzle Inventor to help programmers implement swizzle programs, by writing program sketches that omit swizzles and delegating their creation to an automatic synthesizer. Our synthesis algorithm scales to real-world programs, allowing us to invent new GPU kernels for stencil computations, matrix transposition, and a finite field multiplication algorithm (used in cryptographic applications). The synthesized 2D convolution and finite field multiplication kernels are on average 1.5--3.2x and 1.1--1.7x faster, respectively, than expert-optimized CUDA kernels.

We present a new SMT-based, probabilistic, syntax-guided method to discover numerical inductive invariants. The core idea is to initialize frequency distributions from the program's source code, then repeatedly sample lemmas from those distributions, and terminate when the conjunction of learned lemmas becomes a safe invariant. The sampling process gets further optimized by priority distributions fine-tuned after each positive and negative sample. The stochastic nature of this approach admits simple, asynchronous parallelization. We implemented and evaluated this approach in a tool called FreqHorn which shows competitive performance on well-known linear and some non-linear programs.

A significant focus in the United States recently has been to increase engagement and interest in STEM curricula, particularly among girls and underrepresented minorities [3]. In this work, we take an approach to teaching and learning that supports flexibility, experimentation, and play with technology. With this approach, we aim to make STEM curricula more comfortable and engaging for all types of children and teens, with a particular emphasis on lower socio-economic status female students. We designed and tested a computing course for middle school girls, and this work resulted in three best practices: hands-on work incorporating creativity through crafts into engineering and computing, the frequent presence of an audience to motivate engagement, and engineering-focused individual roles structuring group work. Pre- and post-surveys and exit interviews revealed significant changes in attitudes and an enthusiasm for engineering projects and careers as a result of participation in the course.

Interventions to support children with cognitive and social developmental disabilities often include visual elements. Use of visual artifacts has been shown to increase the communication and understanding levels of children with disabilities. We describe a research agenda for expanding these capabilities using interactive, collaborative and intelligent systems.

In this paper we report on the results of a large scale user survey investigating the status setting and interruption management behavior of mobile instant messaging (IM) users with existing systems. The motivation for this study was to inform the design of interface tools that support users by setting contextually appropriate awareness messages. Our results demonstrate that many desktop IM practices have been appropriated by mobile laptop users, but in the face of increasingly situated computer usage and an "always online" culture, several frictions are emerging between desktop and mobile practices. We find that common assumptions about IM users and the established awareness cues are failing and users are frequently embarrassed and interrupted with negative and sometimes threatening consequences.