Merge conflicts often arise when developers integrate changes from different software branches. The conflicts can result from overlapping edits in programs (i.e., textual conflicts) or cause build and test errors (i.e., build and test conflicts). They degrade software quality and hinder programmer productivity. While several tools detect build conflicts, few offer meaningful support for resolving them. To overcome limitations of existing tools, we introduce BuCoR (Build Conflict Resolver), a new conflict resolver. BuCoR first detects conflicts by comparing three versions related to a merging scenario: base b, left l, and right r. To resolve conflicts, it employs two complementary strategies: example-based transformation (BuCoR-E) and rule-based transformation (BuCoR-R). BuCoR-R applies predefined rules to resolve conflicts in frequently suggested or conventional ways. BuCoR-E mines branch versions (l and r) for exemplar edits applied to fix related build errors. From these examples, it infers and generalizes program transformation patterns to resolve conflicts in project-specific or unconventional ways. We evaluated BuCoR on 88 real-world build conflicts spanning 21 distinct conflict types. BuCoR generated at least one solution for 65 cases and correctly resolved 34 conflicts. We observed that this hybrid approach–combining context-aware, example-based learning with structured, rule-based resolution–can effectively help resolve conflicts. Our research sheds light on future directions for more intelligent and automated merge tools.

In collaborative software development, program merging is the mechanism to integrate changes from multiple programmers. Merge algorithms in modern version control systems report a conflict when changes interfere textually. Merge conflicts require manual intervention and frequently stall modern continuous integration pipelines. Prior work found that, although costly, a large majority of resolutions involve re-arranging text without writing any new code. Inspired by this observation we propose the first data-driven approach to resolve merge conflicts with a machine learning model. We realize our approach in a tool DeepMerge that uses a novel combination of (i) an edit-aware embedding of merge inputs and (ii) a variation of pointer networks, to construct resolutions from input segments. We also propose an algorithm to localize manual resolutions in a resolved file and employ it to curate a ground-truth dataset comprising 8,719 non-trivial resolutions in JavaScript programs. Our evaluation shows that, on a held out test set, DeepMerge can predict correct resolutions for 37% of non-trivial merges, compared to only 4% by a state-of-the-art semistructured merge technique. Furthermore, on the subset of merges with upto 3 lines (comprising 24% of the total dataset), DeepMerge can predict correct resolutions with 78% accuracy.

Machine learning models are increasingly being trained across multiple GPUs and servers. In this setting, data is transferred between GPUs using communication collectives such as AlltoAll and AllReduce, which can become a significant bottleneck in training large models. Thus, it is important to use efficient algorithms for collective communication. We develop TACCL, a tool that enables algorithm designers to guide a synthesizer into automatically generating algorithms for a given hardware configuration and communication collective. TACCL uses a novel communication sketch abstraction to get crucial information from the designer to significantly reduce the search space and guide the synthesizer towards better algorithms. TACCL also uses a novel encoding of the problem that allows it to scale beyond single-node topologies. We use TACCL to synthesize algorithms for three collectives and two hardware topologies: DGX-2 and NDv2. We demonstrate that the algorithms synthesized by TACCL outperform the Nvidia Collective Communication Library (NCCL) by up to 6.7x. We also show that TACCL can speed up end-to-end training of Transformer-XL and BERT models by 11%--2.3x for different batch sizes.

Testing is widely recognized as an important stage of the software development lifecycle. Effective software testing can provide benefits such as bug finding, preventing regressions, and documentation. In terms of documentation, unit tests express a unit's intended functionality, as conceived by the developer. A test oracle, typically expressed as an condition, documents the intended behavior of a unit under a given test prefix. Synthesizing a functional test oracle is a challenging problem, as it must capture the intended functionality rather than the implemented functionality. In this paper, we propose TOGA (a neural method for Test Oracle GenerAtion), a unified transformer-based neural approach to infer both exceptional and assertion test oracles based on the context of the focal method. Our approach can handle units with ambiguous or missing documentation, and even units with a missing implementation. We evaluate our approach on both oracle inference accuracy and functional bug-finding. Our technique improves accuracy by 33% over existing oracle inference approaches, achieving 96% over-all accuracy on a held out test dataset. Furthermore, we show that when integrated with a automated test generation tool (EvoSuite), our approach finds 57 real world bugs in large-scale Java programs, including 30 bugs that are not found by any other automated testing method in our evaluation.

Recent trends towards large machine learning models require both training and inference tasks to be distributed. Considering the huge cost of training these models, it is imperative to unlock optimizations in computation and communication to obtain best performance. However, the current logical separation between computation and communication kernels in machine learning frameworks misses optimization opportunities across this barrier. Breaking this abstraction can provide many optimizations to improve the performance of distributed workloads. However, manually applying these optimizations requires modifying the underlying computation and communication libraries for each scenario, which is both time consuming and error-prone. Therefore, we present CoCoNet, which contains (i) a domain specific language to express a distributed machine learning program in the form of computation and communication operations, (ii) a set of semantics preserving transformations to optimize the program, and (iii) a compiler to generate jointly optimized communication and computation GPU kernels. Providing both computation and communication as first class constructs allows users to work on a high-level abstraction and apply powerful optimizations, such as fusion or overlapping of communication and computation. CoCoNet enabled us to optimize data-, model- and pipeline-parallel workloads in large language models with only a few lines of code. Our experiments show that CoCoNet significantly outperforms state-of-the-art distributed machine learning implementations.

Collaborative software development is an integral part of the modern software development life cycle, essential to the success of large-scale software projects. When multiple developers make concurrent changes around the same lines of code, a merge conflict may occur. Such conflicts stall pull requests and continuous integration pipelines for hours to several days, seriously hurting developer productivity. To address this problem, we introduce MergeBERT, a novel neural program merge framework based on token-level three-way differencing and a transformer encoder model. By exploiting the restricted nature of merge conflict resolutions, we reformulate the task of generating the resolution sequence as a classification task over a set of primitive merge patterns extracted from real-world merge commit data. Our model achieves 63–68% accuracy for merge resolution synthesis, yielding nearly a 3× performance improvement over existing semi-structured, and 2× improvement over neural program merge tools. Finally, we demonstrate that MergeBERT is sufficiently flexible to work with source code files in Java, JavaScript, TypeScript, and C# programming languages. To measure the practical use of MergeBERT, we conduct a user study to evaluate MergeBERT suggestions with 25 developers from large OSS projects on 122 real-world conflicts they encountered. Results suggest that in practice, MergeBERT resolutions would be accepted at a higher rate than estimated by automatic metrics for precision and accuracy. Additionally, we use participant feedback to identify future avenues for improvement of MergeBERT.

Developing models that can automatically generate detailed code explanation can greatly benefit software maintenance and programming education. However, existing code-to-text generation models often produce only high-level summaries of code that do not capture implementation-level choices essential for these scenarios. To fill in this gap, we propose the code explanation generation task. We first conducted a human study to identify the criteria for high-quality explanatory docstring for code. Based on that, we collected and refined a large-scale code docstring corpus and formulated automatic evaluation metrics that best match human assessments. Finally, we present a multi-stage fine-tuning strategy and baseline models for the task. Our experiments show that (1) our refined training dataset lets models achieve better performance in the explanation generation tasks compared to larger unrefined data (15x larger), and (2) fine-tuned models can generate well-structured long docstrings comparable to human-written ones. We envision our training dataset, human-evaluation protocol, recommended metrics, and fine-tuning strategy can boost future code explanation research. The code and annotated data are available at https://github.com/subercui/CodeExp.

Program merging is standard practice when developers integrate their individual changes to a common code base. When the merge algorithm fails, this is called a merge conflict. The conflict either manifests as a textual merge conflict where the merge fails to produce code, or as a semantic merge conflict where the merged code results in compiler errors or broken tests. Resolving these conflicts for large code projects is expensive because it requires developers to manually identify the sources of conflicts and correct them. In this paper, we explore the feasibility of automatically repairing merge conflicts (both textual and semantic) using k-shot learning with pre-trained large neural language models (LM) such as GPT-3. One of the challenges in leveraging such language models is fitting the examples and the queries within a small prompt (2048 tokens). We evaluate LMs and k-shot learning for both textual and semantic merge conflicts for Microsoft Edge. Our results are mixed: on one-hand, LMs provide the state-of-the-art (SOTA) performance on semantic merge conflict resolution for Edge compared to earlier symbolic approaches; on the other hand, LMs do not yet obviate the benefits of special purpose domain-specific languages (DSL) for restricted patterns for program synthesis.

Modern deep learning workloads run on distributed hardware and are difficult to optimize -- data, model, and pipeline parallelism require a developer to thoughtfully restructure their workload around optimized computation and communication kernels in libraries such as cuBLAS and NCCL. The logical separation between computation and communication leaves performance on the table with missed optimization opportunities across abstraction boundaries. To explore these opportunities, this paper presents CoCoNet, which consists of a compute language to express programs with both computation and communication, a scheduling language to apply transformations on such programs, and a compiler to generate high performance kernels. Providing both computation and communication as first class constructs enables new optimizations, such as overlapping or fusion of communication with computation. CoCoNet allowed us to optimize several data, model and pipeline parallel workloads in existing deep learning systems with very few lines of code. We show significant improvements after integrating novel CoCoNet generated kernels.

Collective communication algorithms are an important component of distributed computation. Indeed, in the case of deep-learning, collective communication is the Amdahl's bottleneck of data-parallel training. This paper introduces SCCL (for Synthesized Collective Communication Library), a systematic approach to synthesize collective communication algorithms that are explicitly tailored to a particular hardware topology. SCCL synthesizes algorithms along the Pareto-frontier spanning from latency-optimal to bandwidth-optimal implementations of a collective. The paper demonstrates how to encode SCCL's synthesis as a quantifier-free SMT formula which can be discharged to a theorem prover. We further demonstrate how to scale our synthesis by exploiting symmetries in topologies and collectives. We synthesize and introduce novel latency and bandwidth optimal algorithms not seen in the literature on two popular hardware topologies. We also show how SCCL efficiently lowers algorithms to implementations on two hardware architectures (NVIDIA and AMD) and demonstrate competitive performance with hand optimized collective communication libraries.

Large ML models and datasets have necessitated the use of multi-GPU systems for distributed model training. To harness the power offered by multi-GPU systems, it is critical to eliminate bottlenecks in inter-GPU communication — a problem made challenging by the heterogeneous nature of interconnects. In this work, we present TACCL, a synthesizer for collective communication primitives for large-scale multi-GPU systems. TACCL encodes a profiled topology and input size into a synthesis problem to generate optimized communication algorithms. TACCL is built on top of the standard NVIDIA Collective Communication Library (NCCL), allowing it to be a drop-in replacement for GPU communication in frameworks like PyTorch with minimal changes. TACCL generates algorithms for communication primitives like ALLGATHER, ALLTOALL, and ALLREDUCE that are up to 3× faster than NCCL. Using TACCL’s algorithms speeds up the end-to-end training of an internal mixture of experts model by 17%. By decomposing the optimization problem into parts and leveraging the symmetry in multi-GPU topologies, TACCL synthesizes collectives for up-to 80-GPUs in less than 3 minutes, at least two orders of magnitude faster than other synthesis-based state-of-the-art collective communication libraries.

Given the rise in popularity of social tagging systems, it seems only natural to ask how efficient is the organically evolved tagging vocabulary in describing underlying document objects? Does this distributed process really provide a way to circumnavigate the traditional "vocabulary problem" with ontology? We analyze a social tagging site, namely del.icio.us, with information theory in order to evaluate the efficiency of this social tagging site for encoding navigation paths to information sources. We show that information theory provides a natural and interesting way to understand this efficiency - or the descriptive, encoding power of tags. Our results indicate the efficiency of tags appears to be waning. We discuss the implications of our findings and provide insight into how our methods can be used to design more usable social tagging software.

Testing is widely recognized as an important stage of the software development lifecycle. Effective software testing can provide benefits such as documentation, bug finding, and preventing regressions. In particular, unit tests document a unit's \textit{intended} functionality. A \textit{test oracle}, typically expressed as an condition, documents the intended behavior of the unit under a given test prefix. Synthesizing a functional test oracle is a challenging problem, as it has to capture the intended functionality and not the implemented functionality. In this paper, we propose \toolName{} (\underline{N}eural \underline{U}nit \underline{T}est \underline{S}uggestions), a unified transformer-based neural approach to infer both exceptional and assertion test oracles based on the context of the focal method. Our approach can handle units with ambiguous or missing documentations, and even units with a missing implementation. We evaluate our approach on both oracle inference accuracy and functional bug-finding. Our technique improves accuracy by 33\% over existing oracle inference approaches, achieving 96\% overall accuracy on a held out test dataset. Furthermore, we show that when integrated with a automated test generation tool (EvoSuite), our approach finds 54 real world bugs in large-scale Java programs, including \DISTINCTBUGS{} bugs that are not found by any other automated testing method in our evaluation.

Many applications today, such as NLP, network analysis, and code analysis, rely on semantically embedding objects into low-dimensional fixed-length vectors. Such embeddings naturally provide a way to perform useful downstream tasks, such as identifying relations among objects or predicting objects for a given context, etc. Unfortunately, the training necessary for accurate embeddings is usually computationally intensive and requires processing large amounts of data. Furthermore, distributing this training is challenging. Most embedding training uses stochastic gradient descent (SGD), an "inherently" sequential algorithm. Prior approaches to parallelizing SGD do not honor these dependencies and thus potentially suffer poor convergence. This paper presents a distributed training framework for a class of applications that use Skip-gram-like models to generate embeddings. We call this class Any2Vec and it includes Word2Vec, DeepWalk, and Node2Vec among others. We first formulate Any2Vec training algorithm as a graph application and leverage the state-of-the-art distributed graph analytics framework, D-Galois. We adapt D-Galois to support dynamic graph generation and repartitioning, and incorporate novel communication optimizations. Finally, we introduce a novel way to combine gradients during distributed training to prevent accuracy loss. We show that our framework, called GraphAny2Vec, matches on a cluster of 32 hosts the accuracy of the state-of-the-art shared-memory implementations of Word2Vec and Vertex2Vec on 1 host, and gives a geo-mean speedup of 12x and 5x respectively. Furthermore, GraphAny2Vec is on average 2x faster than the state-of-the-art distributed Word2Vec implementation, DMTK, on 32 hosts. We also show the superiority of our Gradient Combiner independent of GraphAny2Vec by incorporating it in DMTK, which raises its accuracy by > 30%.

Stochastic gradient descent (SGD) is an inherently sequential training algorithm--computing the gradient at batch $i$ depends on the model parameters learned from batch $i-1$. Prior approaches that break this dependence do not honor them (e.g., sum the gradients for each batch, which is not what sequential SGD would do) and thus potentially suffer from poor convergence. This paper introduces a novel method to combine gradients called Adasum (for adaptive sum) that converges faster than prior work. Adasum is easy to implement, almost as efficient as simply summing gradients, and is integrated into the open-source toolkit Horovod. This paper first provides a formal justification for Adasum and then empirically demonstrates Adasum is more accurate than prior gradient accumulation methods. It then introduces a series of case-studies to show Adasum works with multiple frameworks, (TensorFlow and PyTorch), scales multiple optimizers (Momentum-SGD, Adam, and LAMB) to larger batch-sizes while still giving good downstream accuracy. Finally, it proves that Adasum converges. To summarize, Adasum scales Momentum-SGD on the MLPerf Resnet50 benchmark to 64K examples before communication (no MLPerf v0.5 entry converged with more than 16K), the Adam optimizer to 64K examples before communication on BERT-LARGE (prior work showed Adam stopped scaling at 16K), and the LAMB optimizer to 128K before communication on BERT-LARGE (prior work used 64K), all while maintaining downstream accuracy metrics. Finally, if a user does not need to scale, we show LAMB with Adasum on BERT-LARGE converges in 30% fewer steps than the baseline.

Fully Homomorphic Encryption (FHE) refers to a set of encryption schemes that allow computations on encrypted data without requiring a secret key. Recent cryptographic advances have pushed FHE into the realm of practical applications. However, programming these applications remains a huge challenge, as it requires cryptographic domain expertise to ensure correctness, security, and performance. CHET is a domain-specific optimizing compiler designed to make the task of programming FHE applications easier. Motivated by the need to perform neural network inference on encrypted medical and financial data, CHET supports a domain-specific language for specifying tensor circuits. It automates many of the laborious and error prone tasks of encoding such circuits homomorphically, including encryption parameter selection to guarantee security and accuracy of the computation, determining efficient tensor layouts, and performing scheme-specific optimizations. Our evaluation on a collection of popular neural networks shows that CHET generates homomorphic circuits that outperform expert-tuned circuits and makes it easy to switch across different encryption schemes. We demonstrate its scalability by evaluating it on a version of SqueezeNet, which to the best of our knowledge, is the deepest neural network to be evaluated homomorphically.

Multilingual data-parallel pipelines, such as Microsoft's Scope and Apache Spark, are widely used in real-world analytical tasks. While the involvement of multiple languages (often including both managed and native languages) provides much convenience in data manipulation and transformation, it comes at a performance cost --- managed languages need a managed runtime, incurring much overhead. In addition, each switch from a managed to a native runtime (and vice versa) requires marshalling or unmarshalling of an ocean of data objects, taking a large fraction of the execution time. This paper presents Niijima, an optimizing compiler for Microsoft's Scope/Cosmos, which can consolidate C#-based user-defined operators (UDOs) across SQL statements, thereby reducing the number of dataflow vertices that require the managed runtime, and thus the amount of C# computations and the data marshalling cost. We demonstrate that Niijima has reduced job latency by an average of 24% and up to 3.3x, on a series of production jobs.

Stochastic gradient descent (SGD) is a well-known method for regression and classification tasks. However, it is an inherently sequential algorithm - at each step, the processing of the current example depends on the parameters learned from previous examples. Prior approaches to parallelizing linear learners using SGD, such as Hogwild! and AllReduce, do not honor these dependencies across threads and thus can potentially suffer poor convergence rates and/or poor scalability. This paper proposes SymSGD, a parallel SGD algorithm that, to a first-order approximation, retains the sequential semantics of SGD. Each thread learns a local model in addition to a model combiner, which allows local models to be combined to produce the same result as what a sequential SGD would have produced. This paper evaluates SymSGD's accuracy and performance on 6 datasets on a shared-memory machine shows up-to 11x speedup over our heavily optimized sequential baseline on 16 cores and 2.2x, on average, faster than Hogwild!.

Fully Homomorphic Encryption (FHE) refers to a set of encryption schemes that allow computations to be applied directly on encrypted data without requiring a secret key. This enables novel application scenarios where a client can safely offload storage and computation to a third-party cloud provider without having to trust the software and the hardware vendors with the decryption keys. Recent advances in both FHE schemes and implementations have moved such applications from theoretical possibilities into the realm of practicalities. This paper proposes a compact and well-reasoned interface called the Homomorphic Instruction Set Architecture (HISA) for developing FHE applications. Just as the hardware ISA interface enabled hardware advances to proceed independent of software advances in the compiler and language runtimes, HISA decouples compiler optimizations and runtimes for supporting FHE applications from advancements in the underlying FHE schemes. This paper demonstrates the capabilities of HISA by building an end-to-end software stack for evaluating neural network models on encrypted data. Our stack includes an end-to-end compiler, runtime, and a set of optimizations. Our approach shows generated code, on a set of popular neural network architectures, is faster than hand-optimized implementations.

Building scalable big data programs currently requires programmers to combine relational (SQL) with non-relational code (Java, C#, Scala). Relational code is declarative a program describes what the computation is and the compiler decides how to distribute the program. SQL query optimization has enjoyed a rich and fruitful history, however, most research and commercial optimization engines treat non-relational code as a black-box and thus are unable to optimize it. This paper empirically studies over 3 million SCOPE programs across five data centers within Microsoft and finds programs with non-relational code take between 45-70% of data center CPU time. We further explore the potential for SCOPE optimization by generating more native code from the non-relational part. Finally, we present 6 case studies showing that triggering more generation of native code in these jobs yields significant performance improvement: optimizing just one portion resulted in as much as 25% improvement for an entire program.