EVA: An Encrypted Vector Arithmetic Language and Compiler for Efficient Homomorphic Computation

Fully-Homomorphic Encryption (FHE) offers powerful capabilities by enabling secure offloading of both storage and computation, and recent innovations in schemes and implementations have made it all the more attractive. At the same time, FHE is notoriously hard to use with a very constrained programming model, a very unusual performance profile, and many cryptographic constraints. Existing compilers for FHE either target simpler but less efficient FHE schemes or only support specific domains where they can rely on expert-provided high-level runtimes to hide complications. This paper presents a new FHE language called Encrypted Vector Arithmetic (EVA), which includes an optimizing compiler that generates correct and secure FHE programs, while hiding all the complexities of the target FHE scheme. Bolstered by our optimizing compiler, programmers can develop efficient general-purpose FHE applications directly in EVA. For example, we have developed image processing applications using EVA, with a very few lines of code. EVA is designed to also work as an intermediate representation that can be a target for compiling higher-level domain-specific languages. To demonstrate this, we have re-targeted CHET, an existing domain-specific compiler for neural network inference, onto EVA. Due to the novel optimizations in EVA, its programs are on average 5.3x faster than those generated by CHET. We believe that EVA would enable a wider adoption of FHE by making it easier to develop FHE applications and domain-specific FHE compilers.

Lightweight Modular Staging and Embedded Compilers

Tiark Rompf

Programs expressed in a high-level programming language need to be translated to a low-level machine dialect for execution. This translation is usually accomplished by a compiler, which is able to translate any legal program to equivalent low-level code. But for individual source programs, automatic translation does not always deliver good results: Software engineering practice demands generalization and abstraction, whereas high performance demands specialization and concretization. These goals are at odds, and compilers can only rarely translate expressive high-level programs tomodern hardware platforms in a way that makes best use of the available resources. Explicit program generation is a promising alternative to fully automatic translation. Instead of writing down the program and relying on a compiler for translation, developers write a program generator, which produces a specialized, efficient, low-level program as its output. However, developing high-quality program generators requires a very large effort that is often hard to amortize. In this thesis, we propose a hybrid design: Integrate compilers into programs so that programs can take control of the translation process, but rely on libraries of common compiler functionality for help. We present Lightweight Modular Staging (LMS), a generative programming approach that lowers the development effort significantly. LMS combines program generator logic with the generated code in a single program, using only types to distinguish the two stages of execution. Through extensive use of component technology, LMS makes a reusable and extensible compiler framework available at the library level, allowing programmers to tightly integrate domain-specific abstractions and optimizations into the generation process, with common generic optimizations provided by the framework. Compared to previous work on programgeneration, a key aspect of our design is the use of staging not only as a front-end, but also as a way to implement internal compiler passes and optimizations, many of which can be combined into powerful joint simplification passes. LMS is well suited to develop embedded domain specific languages (DSLs) and has been used to develop powerful performance-oriented DSLs for demanding domains such as machine learning, with code generation for heterogeneous platforms including GPUs. LMS has also been used to generate SQL for embedded database queries and JavaScript for web applications.

EPFL2012

Program Analysis and Compilation Techniques for Speeding up Transactional Database Workloads

Mohammad Dashti Rahmat Abadi

There is a trend towards increased specialization of data management software for performance reasons. The improved performance not only leads to a more efficient usage of the underlying hardware and cuts the operation costs of the system, but also is a game-changing competitive advantage for many emerging application domains such as high-frequency algorithmic trading, clickstream analysis, infrastructure monitoring, fraud detection, and online advertising to name a few. In this thesis, we study the automatic specialization and optimization of database application programs -- sequences of queries and updates, augmented with control flow constructs as they appear in database scripts, user-defined functions (UDFs), transactional workloads and triggers in languages such as PL/SQL. We propose to build online transaction processing (OLTP) systems around a modern compiler infrastructure. We show how to build an optimizing compiler for transaction programs using generative programming and state-of-the-art compiler technology, and present techniques for aggressive code inlining, fusion, deforestation, and data structure specialization in the domain of relational transaction programs. We also identify and explore the key optimizations that can be applied in this domain. In addition, we study the advantage of using program dependency analysis and restructuring to enable the concurrency control algorithms to achieve higher performance. Traditionally, optimistic concurrency control algorithms, such as optimistic Multi-Version Concurrency Control (MVCC), avoid blocking concurrent transactions at the cost of having a validation phase. Upon failure in the validation phase, the transaction is usually aborted and restarted from scratch. The "abort and restart" approach becomes a performance bottleneck for use cases with high contention objects or long running transactions. In addition, restarting from scratch creates a negative feedback loop in the system, because the system incurs additional overhead that may create even more conflicts. However, using the dependency information inside the transaction programs, we propose a novel transaction repair approach for in-memory databases. This low overhead approach summarizes the transaction programs in the form of a dependency graph. The dependency graph also contains the constructs used in the validation phase of the MVCC algorithm. Then, when encountering conflicts among transactions, our mechanism quickly detects the conflict locations in the program and partially re-executes the conflicting transactions. This approach maximizes the reuse of the computations done in the first execution round and increases the transaction processing throughput. We evaluate the proposed ideas and techniques in the thesis on some popular benchmarks such as TPC-C and modified versions of TPC-H and TPC-E, as well as other micro-benchmarks. We show that applying these techniques leads to 2x-100x performance improvement in many use cases. Besides, by selectively disabling some of the optimizations in the compiler, we derive a clinical and precise way of obtaining insight into their individual performance contributions.

EPFL2017

Lightweight Modular Staging and Embedded Compilers

Tiark Rompf

EPFL2012

Program Analysis and Compilation Techniques for Speeding up Transactional Database Workloads

Mohammad Dashti Rahmat Abadi

EPFL2017