This paper presents ACPO: An AI-Enabled Compiler Framework; a novel framework that provides LLVM with simple and comprehensive tools to enable employing ML models for different optimization passes. We showcase a couple of use cases of ACPO by ML-enabling the Loop Unroll (LU) and Function Inlining (FI) passes and experimental results reveal that by including both models, ACPO can provide a combined speedup of 2.4% on Cbench when compared with LLVM’s O3.

The IFDS algorithm can be both memory- and compute-intensive for large programs as it needs to store a huge amount of path edges in memory and process them until a fixed point. In general, an IFDS-based data-flow analysis, such as taint analysis, aims to discover only the data-flow facts at some program points. Maintaining a huge amount of path edges (with many visited only once) wastes memory resources, and consequently, reduces its scalability and efficiency (due to frequent re-hashings for the path-edge data structure used).

The extensive use of libraries in modern software impedes the scalability of pointer analysis. To address this issue, library summarization can be beneficial, but only if the resulting summary-based pointer analysis is faster without sacrificing much precision in the application code. However, currently, no library summarization approaches exist that meet this design objective. This paper presents a novel approach that solves this problem by using k-object-sensitive pointer analysis, k-obj, for Java. The approach involves applying k-obj, along with a set of summary-based inference rules, to generate a k-object-sensitive library summary. By replacing the program's library with this summary and applying k-obj, the efficiency of the program can be significantly improved while maintaining nearly the same or better precision in the application code. We validate our approach with an implementation in Soot and an evaluation using representative Java programs.

Object-sensitive pointer analysis (denotedkobjunder$k$-limiting) for an object-oriented program can be accelerated if context-sensitivity can be selectively applied to only some precision-critical variables/objects in a program. Existing pre-analyses for making such selections, which are performed as whole-program analyses to a program, are developed based on two broad approaches. One approach preserves the precision of object-sensitive pointer analysis but achieves limited speedups by reasoning about all the possible value flows in the program conservatively, while the other approach achieves greater speedups but sacrifices precision (often unduly) by examining only some but not all the value flows in the program heuristically. In this paper, we introduce a new pre-analysis approach,Turner$^{\mathcal{m}}$(where$\mathcal {m}$stands for modularity), that represents a sweet spot between these two existing ones, as it is designed to enablekobjto run significantly faster than the former approach and achieve significantly better precision than the latter approach.Turner$^{\mathcal{m}}$is simple, lightweight yet effective due to two novel aspects in its design. First, we exploit a key observation that some precision-uncritical objects in the program can be approximated based on the object-containment relationship pre-established (from Andersen's analysis). In practice, this approximation introduces only a small degree of imprecision intokobj. Second, leveraging this initial approximation, we apply a novel object reachability analysis to the program by pre-analyzing its methods according to a reverse topological order of its call graph. When pre-analyzing each method, we make use of a simple DFA (Deterministic Finite Automaton) to reason about object reachability intra-procedurally from its entry to its exit along all the possible value flows established by its statements to identify its precision-critical variables/objects. In practice, this new modular object reachability analysis, which runs linearly in terms of the number of statements in the program, introduces again only a small loss of precision intokobj. We have validatedTurner$^{\mathcal{m}}$with an open-source implementation inSoot(already publicly available) against the state of the art by using a set of 12 widely used Java benchmarks and applications.

This paper presents MLGOPerf; the first end-to-end framework capable of optimizing performance using LLVM’s ML-Inliner. It employs a secondary ML model to generate rewards used for training a retargeted Reinforcement learning agent, previously used as the primary model by MLGO. It does so by predicting the post-inlining speedup of a function under analysis and it enables a fast training framework for the primary model which otherwise wouldn’t be practical. The experimental results show MLGOPerf is able to gain up to 1.8% with respect to LLVM’s optimization at O3 when trained for performance on SPEC CPU2006. Furthermore, the proposed approach provides up to 26% increased opportunities to autotune code regions for our benchmarks which can be translated into an additional 3.7% speedup value.

Control-Flow Integrity (CFI) techniques focus often on protecting forward edges and assume that backward edges are protected by shadow stacks. However, software-based shadow stacks that can provide performance, security, and compatibility are still hard to obtain, leaving an important security gap on x86-64. In this article, we introduce a simple, efficient, and effective parallel shadow stack design (based on LLVM), FlashStack , for protecting return addresses in single- and multi-threaded programs running under 64-bit Linux on x86-64, with three distinctive features. First, we introduce a novel dual-prologue approach to enable a protected function to thwart the TOCTTOU attacks, which are constructed by Microsoft’s red team and lead to the deprecation of Microsoft’s RFG. Second, we design a new mapping mechanism, Segment+Rsp-S , to allow the parallel shadow stack to be accessed efficiently while satisfying the constraints of arch_prctl() and ASLR in 64-bit Linux. Finally, we introduce a lightweight inspection mechanism, SideChannel-K , to harden FlashStack further by detecting entropy-reduction attacks efficiently and protecting the parallel shadow stack effectively with a 10-ms shuffling policy. Our evaluation on SPEC CPU2006 , Nginx, and Firefox shows that FlashStack can provide high performance, meaningful security, and reasonable compatibility for server- and client-side programs on x86-64.

Shadow stacks play an important role in protecting return addresses to mitigate ROP attacks. Parallel shadow stacks, which shadow the call stack of each thread at the same constant offset for all threads, are known not to support multi-threading well. On the other hand, compact shadow stacks must maintain a separate shadow stack pointer in thread-local storage (TLS) , which can be implemented in terms of a register or the per-thread Thread-Control-Block (TCB) , suffering from poor compatibility in the former or high performance overhead in the latter. In addition, shadow stacks are vulnerable to information disclosure attacks. In this paper, we propose to mitigate ROP attacks for single- and multi-threaded server programs running on general-purpose computing systems by using a novel stack layout, called a buddy stack (referred to as Bustk ), that is highly performant, compatible with existing code, and provides meaningful security. These goals are met due to three novel design aspects in Bustk . First, Bustk places a parallel shadow stack just below a thread’s call stack (as each other’s buddies allocated together), avoiding the need to maintain a separate shadow stack pointer and making it now well-suited for multi-threading. Second, Bustk uses an efficient stack-based thread-local storage mechanism, denoted STK-TLS , to store thread-specific metadata in two TLS sections just below the shadow stack in dual redundancy (as each other’s buddies), so that both can be accessed and updated in a lightweight manner from the call stack pointer rsp alone. Finally, Bustk re-randomizes continuously (on the order of milliseconds) the return addresses on the shadow stack by using a new microsecond-level runtime re-randomization technique, denoted STK-MSR . This mechanism aims to obsolete leaked information, making it extremely unlikely for the attacker to hijack return addresses, particularly against a server program that sits often tens of milliseconds away from the attacker. Our evaluation using web servers, Nginx and Apache Httpd , shows that Bustk works well in terms of performance, compatibility, and security provided, with its parallel shadow stacks incurring acceptable memory overhead for real-world applications and its STK-TLS mechanism costing only two pages per thread. In particular, Bustk can protect the Nginx and Apache servers with an adaptive 1-ms re-randomization policy (without observable overheads when IO is intensive, with about 17,000 requests per second). In addition, we have also evaluated Bustk using other non-server applications, Firefox , Python , LLVM , JDK and SPEC CPU2006 , to demonstrate further the same degree of performance and compatibility provided, but the protection provided for, say, browsers, is weaker (since network-access delays can no longer be assumed).

Object-sensitive pointer analysis for an object-oriented program can be accelerated if context-sensitivity can be selectively applied to some precision-critical variables/objects in the program. Existing pre-analyses, which are performed to make such selections, either preserve precision but achieve limited speedups by reasoning about all the possible value flows in the program conservatively or achieve greater speedups but sacrifice precision (often unduly) by examining only some but not all the value flows in the program heuristically. In this paper, we introduce a new approach, named Turner, that represents a sweet spot between the two existing ones, as it is designed to enable object-sensitive pointer analysis to run significantly faster than the former approach and achieve significantly better precision than the latter approach. Turner is simple, lightweight yet effective due to two novel aspects in its design. First, we exploit a key observation that some precision-uncritical objects can be approximated based on the object-containment relationship pre-established (by applying Andersen’s analysis). This approximation introduces a small degree yet the only source of imprecision into Turner. Second, leveraging this initial approximation, we introduce a simple DFA to reason about object reachability for a method intra-procedurally from its entry to its exit along all the possible value flows established by its statements to finalize its precision-critical variables/objects identified. We have validated Turner with an implementation in Soot against the state of the art using a set of 12 popular Java benchmarks and applications.

Feedback-driven optimization (FDO) is an important component in mainstream compilers. By allowing the compiler to reoptimize the program based on some profiles of the program's dynamic behaviors, it often enhances the quality of the generated code substantially. A barrier for using FDO is that it often requires many training runs to collect enough profiles to amortize the sensitivity of program optimizations to program input changes. Various sampling techniques have been explored to alleviate this time-consuming process. However, the lowered profile accuracy caused by sampling often hurts the benefits of FDO. This article gives the first systematic study in how sampling rates affect the accuracy of collected profiles and how the accuracy correlates with the usefulness of the profile for modern FDO. Studying basic block and edge profiles for FDO in two mature compilers reveals several counterintuitive observations, one of which is that profiling accuracy does not strongly correlate with the benefits of the FDO. A detailed analysis identifies three types of sampling-caused errors that critically impair the quality of the profiles for FDO. It then introduces a simple way to rectify profiles based on the findings. Experiments demonstrate that the simple rectification fixes most of those critical errors in sampled profiles and significantly enhances the effectiveness of FDO.

Function versioning is an approach to addressing input-sensitivity of program optimizations. A major side effect of it is notable code size increase, which has been hindering its broad applications to large code bases and space-stringent environments. In this paper, we initiate a systematic exploration into the problem, providing answers to some fundamental questions: Given a space constraint, to which function we should apply versioning? How many versions of a function should we include in the final executable? Is the optimal selection feasible to do in polynomial time? This study proves selecting the best set of versions under a space constraint is NP-complete and proposes a heuristic algorithm named CHoGS which yields near optimal results in quadratic time. We implement the algorithm and conduct experiments through the IBM XL compilers. We observe significant performance enhancement with only slight code size increase; the results from CHoGS show factors of higher space efficiency than those from traditional hotness-based methods.

Pipelining is necessary for efficient do-across parallelism but the use is difficult to automate because it requires send-receive analysis and loop blocking in both sender and receiver code. The blocking factor is statically chosen. This paper presents a new interface called delta send-recv. Through compiler and run-time support, it enables dynamic pipelining. In program code, the interface is used to mark the related computation and communication. There is no need to restructure the computation code or compose multiple messages. At run time, the message size is dynamically determined, and multiple pipelines are chained among all tasks that participate in the delta communication. The new system is tested on kernel and reduced NAS benchmarks to show that it simplifies message-passing programming and improves program performance.

Prediction of program dynamic behaviors is fundamental to program optimizations, resource management, and architecture reconfigurations. Most existing predictors are based on locality of program behaviors, subject to some inherent limitations. In this paper, we revisit the design philosophy and systematically explore a second source of clues: statistical correlations between the behavior sequences of different program entities. Concentrated on loops, it examines the correlations' existence, strength, and values in enhancing the design of program behavior predictors. It creates the first taxonomy of program behavior sequence patterns. It develops a new form of predictors, named sequence predictors, to effectively translate the correlations into large-scope, proactive predictions of program behavior sequences. It demonstrates the usefulness of the prediction in dynamic version selection and loop importance estimation, showing 19% average speedup on a number of real-world utility applications. By taking scope and timing of behavior prediction as the first-order design objectives, the new approach overcomes limitations of existing program behavior predictors, opening up many new opportunities for runtime optimizations at various layers of computing.

Most of today's processors include vector units that have been designed to speedup single threaded programs. Although vector instructions can deliver high performance, writing vector code in assembly language or using intrinsics in high level languages is a time consuming and error-prone task. The alternative is to automate the process of vectorization by using vectorizing compilers. This paper evaluates how well compilers vectorize a synthetic benchmark consisting of 151 loops, two application from Petascale Application Collaboration Teams (PACT), and eight applications from Media Bench II. We evaluated three compilers: GCC (version 4.7.0), ICC (version 12.0) and XLC (version 11.01). Our results show that despite all the work done in vectorization in the last 40 years 45-71% of the loops in the synthetic benchmark and only a few loops from the real applications are vectorized by the compilers we evaluated.

Many techniques characterize the program working set by the notion of the program footprint, which is the volume of data accessed in a time window. A complete characterization requires measuring data access in all O(n2) windows in an n-element trace. Two recent techniques have significantly reduced the measurement time, but the cost is still too high for real-size workloads. Instead of measuring all footprint sizes, this paper presents a technique for measuring the average footprint size. By confining the analysis to the average rather than the full range, the problem can be solved accurately by a linear-time algorithm. The paper presents the algorithm and evaluates it using the complete suites of 26 SPEC2000 and 29 SPEC2006 benchmarks. The new algorithm is compared against the previously fastest algorithm in both the speed of the measurement and the accuracy of shared-cache performance prediction.

This paper presents a finding and a technique on program behavior prediction. The finding is that surprisingly strong statistical correlations exist among the behaviors of different program components (e.g., loops) and among different types of program level behaviors (e.g., loop trip-counts versus data values). Furthermore, the correlations can be beneficially exploited: They help resolve the proactivity-adaptivity dilemma faced by existing program behavior predictions, making it possible to gain the strengths of both approaches--the large scope and earliness of offline-profiling--based predictions, and the cross-input adaptivity of runtime sampling-based predictions.

This paper describes Memory-Pooling-Assisted Data Splitting (MPADS), a framework that combines data structure splitting with memory pooling --- Although it MPADS may call to mind memory padding, a distintion of this framework is that is does not insert padding. MPADS relies on pointer analysis to ensure that splitting is safe and applicable to type-unsafe language. MPADS makes no assumption about type safety. The analysis can identify cases in which the transformation could lead to incorrect code and thus MPADS abandons those cases.

This article presents Forma , a practical, safe, and automatic data reshaping framework that reorganizes arrays to improve data locality. Forma splits large aggregated data-types into smaller ones to improve data locality. Arrays of these large data types are then replaced by multiple arrays of the smaller types. These new arrays form natural data streams that have smaller memory footprints, better locality, and are more suitable for hardware stream prefetching. Forma consists of a field-sensitive alias analyzer, a data type checker, a portable structure reshaping planner, and an array reshaper. An extensive experimental study compares different data reshaping strategies in two dimensions: (1) how the data structure is split into smaller ones ( maximal partition × frequency-based partition × affinity-based partition ); and (2) how partitioned arrays are linked to preserve program semantics ( address arithmetic-based reshaping × pointer-based reshaping ). This study exposes important characteristics of array reshaping. First, a practical data reshaper needs not only an inter-procedural analysis but also a data-type checker to make sure that array reshaping is safe. Second, the performance improvement due to array reshaping can be dramatic: standard benchmarks can run up to 2.1 times faster after array reshaping. Array reshaping may also result in some performance degradation for certain benchmarks. An extensive micro-architecture-level performance study identifies the causes for this degradation. Third, the seemingly naive maximal partition achieves best or close-to-best performance in the benchmarks studied. This article presents an analysis that explains this surprising result. Finally, address-arithmetic-based reshaping always performs better than its pointer-based counterpart.

Previous studies have shown that array regrouping and structure splitting significantly improve data locality. The most effective technique relies on profiling every access to every data element. The high overhead impedes its adoption in a general compiler, In this paper, we show that for array regrouping in scientific programs, the overhead is not needed since the same benefit can be obtained by pure program analysis.We present an interprocedural analysis technique for array regrouping. For each global array, the analysis summarizes the access pattern by access-frequency vectors and then groups arrays with similar vectors. The analysis is context sensitive, so it tracks the exact array access. For each loop or function call, it uses two methods to estimate the frequency of the execution. The first is symbolic analysis in the compiler. The second is lightweight profiling of the code. The same interprocedural analysis is used to cumulate the overall execution frequency by considering the calling context. We implemented a prototype of both the compiler and the profiling analysis in the IBM® compiler, evaluated array regrouping on the entire set of SPEC CPU2000 FORTRAN benchmarks, and compared different analysis methods. The pure compiler-based array regrouping improves the performance for the majority of programs, leaving little room for improvement by code or data profiling.

A brief overview is presented of the parallel abstract machine developed for the RAP/LOP parallel execution model. A description is also given of its optimized parallel compiler. The main features of the machine are: (i) the OR-forest description is used to describe the search space of a given problem, not only describing OR- and AND- parallelism explicitly, but also avoiding a class of redundant computations; (ii) coarse-grain parallelism is supported by the granularity-based scheduling policy; (iii) procedure-level and clause-level analysis at compile-time and dynamic simple run-time checks are used to identify independent goals of the body of a clause; and (iv) several optimization and implementation techniques such as improved indexing mechanism and code space reduction are used to increase the machine's efficiency significantly.>