Advanced Compiler Optimizations Research Articles

PROMPT: A Fast and Extensible Memory Profiling Framework

Memory profiling captures programs’ dynamic memory behavior, assisting programmers in debugging, tuning, and enabling advanced compiler optimizations like speculation-based automatic parallelization. As each use case demands its unique program trace summary, various memory profiler types have been developed. Yet, designing practical memory profilers often requires extensive compiler expertise, adeptness in program optimization, and significant implementation effort. This often results in a void where aspirations for fast and robust profilers remain unfulfilled. To bridge this gap, this paper presents PROMPT, a framework for streamlined development of fast memory profilers. With PROMPT, developers need only specify profiling events and define the core profiling logic, bypassing the complexities of custom instrumentation and intricate memory profiling components and optimizations. Two state-of-the-art memory profilers were ported with PROMPT where all features preserved. By focusing on the core profiling logic, the code was reduced by more than 65% and the profiling overhead was improved by 5.3× and 7.1× respectively. To further underscore PROMPT’s impact, a tailored memory profiling workflow was constructed for a sophisticated compiler optimization client. In 570 lines of code, this redesigned workflow satisfies the client’s memory profiling needs while achieving more than 90% reduction in profiling overhead and improved robustness compared to the original profilers.

An Exploration of Approaches to Instruction Pipeline Implementation for Cycle-Accurate Simulators of «Elbrus»

Software simulation is of a big importance during development of processors as they provide access to hardware under development. Cycle-accurate simulators allow software engineers to design and optimize high-performance algorithms and programs with considerations of features and characteristics of processors being in development. This is especially important for architectures, whose performance is mainly achieved by advanced compiler optimizations. One of the core aspects of a cycle-accurate simulator is the way it simulates the pipeline of the target processor. A pipeline model has high impact on an overall structure of a simulator and its potential performance and accuracy. The main goal of this paper is to develop and analyze different approaches to pipeline simulation of “Elbrus” microprocessors, which let us reuse functionality of existing instruction set simulator and achieve good balance of performance and accuracy. We briefly describe features of “Elbrus” microprocessors and specifics of existing instruction set simulator, relevant for cycle-accurate simulation. We make several simple, but general and useful observations about various aspects of pipeline behavior in context of accurate and efficient cycle-accurate simulation of microprocessors. These observations are then used as a basis for justification, development and analysis of the several approaches to the pipeline simulation, described in this paper. We describe four different approaches, starting from simple and obvious one, which is then successively transformed into more advanced ones through several iterations. We analyze limitations of proposed approaches and outline further work.

A relational model of types-and-effects in higher-order concurrent separation logic

Recently we have seen a renewed interest in programming languages that tame the complexity of state and concurrency through refined type systems with more fine-grained control over effects. In addition to simplifying reasoning and eliminating whole classes of bugs, statically tracking effects opens the door to advanced compiler optimizations. In this paper we present a relational model of a type-and-effect system for a higher-order, concurrent program- ming language. The model precisely captures the semantic invariants expressed by the effect annotations. We demonstrate that these invariants are strong enough to prove advanced program transformations, including automatic parallelization of expressions with suitably disjoint effects. The model also supports refinement proofs between abstract data types implementations with different internal data representations, including proofs that fine-grained concurrent algorithms refine their coarse-grained counterparts. This is the first model for such an expressive language that supports both effect-based optimizations and data abstraction. The logical relation is defined in Iris, a state-of-the-art higher-order concurrent separation logic. This greatly simplifies proving well-definedness of the logical relation and also provides us with a powerful logic for reasoning in the model.

The Devil truly is in the detail. A cautionary note on computational determinism: Implications for structural geology numerical codes and interpretation of their results

The widespread use of numerical modeling codes in structural geology, particularly in forward modeling of crustal deformation, has brought valuable insight into many processes such as fault initiation and propagation, fold growth, and orogenic wedge development. Whether commercial, open source, or the result of an individual researcher’s endeavor, such numerical codes increasingly rely on advanced compiler optimizations and/or parallelization techniques. This is done to maximize performance on today’s multicore processors and thus achieve tractable model runtimes. However, such speed and productivity can come at a price. Many optimization and parallelization techniques used do indeed produce remarkable runtime speedups, but at the expense of determinism and reproducibility. Most researchers are only tangentially aware of these issues. I have aimed to explain why numerical model runs with identical sets of initial and boundary conditions can potentially produce different results and thus, sensu stricto, be nondeterministic. I have developed some solutions and guidelines for maintaining determinism and evaluated these issues using a discrete element code of upper crustal deformation. Implications for the interpretation and use of such model results in structural geology were also evaluated.

Special issue on parallelism

The articles presented in our Special Issue on parallel processing on the supercomputing scale reflect, to some extent, splits in the community developing these machines. There are several schools ...