Amey Karkare

Heap reference analysis using access graphs

Despite significant progress in the theory and practice of program analysis, analyzing properties of heap data has not reached the same level of maturity as the analysis of static and stack data. The spatial and temporal structure of stack and static data is well understood while that of heap data seems arbitrary and is unbounded. We devise bounded representations that summarize properties of the heap data. This summarization is based on the structure of the program that manipulates the heap. The resulting summary representations are certain kinds of graphs called access graphs. The boundedness of these representations and the monotonicity of the operations to manipulate them make it possible to compute them through data flow analysis. An important application that benefits from heap reference analysis is garbage collection, where currently liveness is conservatively approximated by reachability from program variables. As a consequence, current garbage collectors leave a lot of garbage uncollected, a fact that has been confirmed by several empirical studies. We propose the first ever end-to-end static analysis to distinguish live objects from reachable objects. We use this information to make dead objects unreachable by modifying the program. This application is interesting because it requires discovering data flow information representing complex semantics. In particular, we formulate the following new analyses for heap data: liveness, availability, and anticipability and propose solution methods for them. Together, they cover various combinations of directions of analysis (i.e., forward and backward) and confluence of information (i.e. union and intersection). Our analysis can also be used for plugging memory leaks in C/C++ languages.

Program synthesis using natural language

Interacting with computers is a ubiquitous activity for millions of people. Repetitive or specialized tasks often require creation of small, often one-off, programs. End-users struggle with learning and using the myriad of domain-specific languages (DSLs) to effectively accomplish these tasks. We present a general framework for constructing program synthesizers that take natural language (NL) inputs and produce expressions in a target DSL. The framework takes as input a DSL definition and training data consisting of NL/DSL pairs. From these it constructs a synthesizer by learning optimal weights and classifiers (using NLP features) that rank the outputs of a keyword-programming based translation. We applied our framework to three domains: repetitive text editing, an intelligent tutoring system, and flight information queries. On 1200+ English descriptions, the respective synthesizers rank the desired program as the top-1 and top-3 for 80% and 90% descriptions respectively.

A feasibility study of using automated program repair for introductory programming assignments

Despite the fact an intelligent tutoring system for programming (ITSP) education has long attracted interest, its widespread use has been hindered by the difficulty of generating personalized feedback automatically. Meanwhile, automated program repair (APR) is an emerging new technology that automatically fixes software bugs, and it has been shown that APR can fix the bugs of large real-world software. In this paper, we study the feasibility of marrying intelligent programming tutoring and APR. We perform our feasibility study with four state-of-the-art APR tools (GenProg, AE, Angelix, and Prophet), and 661 programs written by the students taking an introductory programming course. We found that when APR tools are used out of the box, only about 30% of the programs in our dataset are repaired. This low repair rate is largely due to the student programs often being significantly incorrect - in contrast, professional software for which APR was successfully applied typically fails only a small portion of tests. To bridge this gap, we adopt in APR a new repair policy akin to the hint generation policy employed in the existing ITSP. This new repair policy admits partial repairs that address part of failing tests, which results in 84% improvement of repair rate. We also performed a user study with 263 novice students and 37 graders, and identified an understudied problem; while novice students do not seem to know how to effectively make use of generated repairs as hints, the graders do seem to gain benefits from repairs.

Automatic Grading and Feedback using Program Repair for Introductory Programming Courses

We present GradeIT, a system that combines the dual objectives of automated grading and program repairing for introductory programming courses (CS1). Syntax errors pose a significant challenge for testcase-based grading as it is difficult to differentiate between a submission that is almost correct and has some minor syntax errors and another submission that is completely off-the-mark. GradeIT also uses program repair to help in grading submissions that do not compile. This enables running testcases on submissions containing minor syntax errors, thereby awarding partial marks for these submissions (which, without repair, do not compile successfully and, hence, do not pass any testcase). Our experiments on 15613 submissions show that GradeIT results are comparable to manual grading by teaching assistants (TAs), and do not suffer from unintentional variability that happens when multiple TAs grade the same assignment. The repairs performed by GradeIT enabled successful compilation of 56% of the submissions having compilation errors, and resulted in an improvement in marks for 11% of these submissions.

Liveness-Based Garbage Collection

Current garbage collectors leave much heap-allocated data uncollected because they preserve data reachable from a root set. However, only live data—a subset of reachable data—need be preserved.

Improving GPU Performance Through Resource Sharing

Graphics Processing Units (GPUs) consisting of Streaming Multiprocessors (SMs) achieve high throughput by running a large number of threads and context switching among them to hide execution latencies. The number of thread blocks, and hence the number of threads that can be launched on an SM, depends on the resource usage--e.g. number of registers, amount of shared memory--of the thread blocks. Since the allocation of threads to an SM is at the thread block granularity, some of the resources may not be used up completely and hence will be wasted. We propose an approach that shares the resources of SM to utilize the wasted resources by launching more thread blocks. We show the effectiveness of our approach for two resources: register sharing, and scratchpad (shared memory) sharing. We further propose optimizations to hide long execution latencies, thus reducing the number of stall cycles. We implemented our approach in GPGPU-Sim simulator and experimentally validated it on 19 applications from 4 different benchmark suites: GPGPU-Sim, Rodinia, CUDA-SDK, and Parboil. We observed that applications that underutilize register resource show a maximum improvement of 24% and an average improvement of 11% with register sharing. Similarly, the applications that underutilize scratchpad resource show a maximum improvement of 30% and an average improvement of 12.5% with scratchpad sharing. The remaining applications, which do not waste any resources, perform similar to the baseline approach.

Compilation error repair: for the student programs, from the student programs

Compile-time errors pose a major learning hurdle for students of introductory programming courses. Compiler error messages, while accurate, are targeted at seasoned programmers, and seem cryptic to beginners. In this work, we address this problem of pedagogically-inspired program repair and report TRACER (Targeted RepAir of Compilation ERrors), a system for performing repairs on compilation errors, aimed at introductory programmers. TRACER invokes a novel combination of tools from programming language theory and deep learning and offers repairs that not only enable successful compilation, but repairs that are very close to those actually performed by students on similar errors. The ability to offer such targeted corrections, rather than just code that compiles, makes TRACER more relevant in offering real-time feedback to students in lab or tutorial sessions, as compared to existing works that merely offer a certain compilation success rate. In an evaluation on 4500 erroneous C programs written by students of a freshman year programming course, TRACER recommends a repair exactly matching the one expected by the student for 68% of the cases, and in 79.27% of the cases, produces a compilable repair. On a further set of 6971 programs that require errors to be fixed on multiple lines, TRACER enjoyed a success rate of 44% compared to the 27% success rate offered by the state-of-the-art technique DeepFix.

Liveness-based garbage collection for lazy languages

We consider the problem of reducing the memory required to run lazy first-order functional programs. Our approach is to analyze programs for liveness of heap-allocated data. The result of the analysis is used to preserve only live data—a subset of reachable data—during garbage collection. The result is an increase in the garbage reclaimed and a reduction in the peak memory requirement of programs. Whereas this technique has already been shown to yield benefits for eager first-order languages, the lack of a statically determinable execution order and the presence of closures pose new challenges for lazy languages. These require changes both in the liveness analysis itself and in the design of the garbage collector. To show the effectiveness of our method, we implemented a copying collector that uses the results of the liveness analysis to preserve live objects, both evaluated and closures. Our experiments confirm that for programs running with a liveness-based garbage collector, there is a significant decrease in peak memory requirements. In addition, a sizable reduction in the number of collections ensures that in spite of using a more complex garbage collector, the execution times of programs running with liveness and reachability-based collectors remain comparable.

Functional SMT solving with Z3 and racket

Satisfiability Modulo Theories (SMT) solvers are powerful tools that can quickly solve complex constraints involving Booleans, integers, first-order logic predicates, lists, and other data types. They have a vast number of potential applications, from constraint solving to program analysis and verification. However, they are so complex to use that their power is inaccessible to all but experts in the field. We present an attempt to make using SMT solvers simpler by integrating the Z3 solver into a host language, Racket. The system defines a programmers interface in Racket that makes it easy to harness the power of Z3 to discover solutions to logical constraints. The interface, although in Racket, retains the structure and brevity of the SMT-LIB format. This system is expected to be useful for a wide variety of applications, from simple constraint solving to writing tools for debugging, verification, and automatic test generation for functional programs.

Precise shape analysis using field sensitivity

We present a static shape analysis technique to infer the shapes of the heap structures created by a program at run time. Our technique is field sensitive in that it uses field information to compute the shapes. The shapes of the heap structures are computed using two components: (a) Boolean functions that capture the shape transitions due to the update of a field in a structure, and (b) through path matrices that store approximate path information between two pointer variables. We classify the shapes as one of Tree, Directed Acyclic Graph (DAG) and Cycle. The novelty of our approach lies in the way we use field information to remember the fields that cause a heap structure to have a particular shape (Tree, DAG or Cycle). This allows us to easily identify the field updates that cause shape transitions from Cycle to DAG, from Cycle to Tree and from DAG to Tree. This makes our analysis more precise as compared to earlier shape analyses that ignore the fields participating in the formation of a shape. We implemented our analysis in GCC as a dynamic plug-in as an interprocedural data-flow analysis and evaluated it on some standard benchmarks against a field-insensitive shape analysis technique as a baseline approach. We are able to achieve significant precision as compared to the baseline analysis (at the cost of increase in analysis time). In particular, we are able to infer more precise shapes for 4 out 7 Olden benchmarks, and never detect more cycles than the baseline analysis. We further suggest enhancements to improve the precision of our analysis under some constraints and to improve the analysis time at the cost of precision.