Both major open-source compilers, GCC and LLVM, have a mature link-time optimization framework usable on most current programs. They are however not free from many performance issues, which prevent them to perform certain analyses and optimizations. We analyze bottlenecks and identify multiple places for improvement, focusing on improving interprocedural points-to analysis. For this purpose, we design a new data structure derived from Bloom filters and use it to significantly improve performance and memory consumption of link-time optimization. Powered by TCPDF (www.tcpdf.org)

MATLAB is an array language, initially popular for rapid prototyping, but is now being in-creasingly used to develop production code for numerical and scientific applications. Typical MATLAB programs have abundant data parallelism. These programs also have control flow dominated scalar regions that have an impact on the program’s execution time. Today’s com-puter systems have tremendous computing power in the form of traditional CPU cores and also throughput-oriented accelerators such as graphics processing units (GPUs). Thus, an approach that maps the control flow dominated regions of a MATLAB program to the CPU and the data parallel regions to the GPU can significantly improve program performance. In this work, we present the design and implementation of MEGHA, a compiler that auto-matically compiles MATLAB programs to enable synergistic execution on heterogeneous pro-cessors. Our solution is fully automated and does not require programmer input for identifying data parallel regions. Our compiler identifies data parallel regions of the program and com-poses them into kernels. The kernel composition step eliminates a number of intermediate arrays which are otherwise required and also reduces the size of the scheduling and mapping problem the compiler needs to solve subsequently. The problem of combining statements into kernels is formulated as a constrained graph clustering problem. Heuristics are presented to map identified kernels to either the CPU or GPU so that kernel execution on the CPU and the GPU happens synergistically, and the amount of data transfer needed is minimized. A heuristic technique to ensure that memory accesses on the CPU exploit locality and those on the GPU are coalesced is also presented. In order to ensure that data transfers required for dependences across basic blocks are performed, we propose a data flow analysis step and an edge-splitting strategy. Thus our compiler automatically handles kernel composition, mapping of kernels to CPU and GPU, scheduling and insertion of required data transfers. Additionally, we address the problem of identifying what variables can coexist in GPU memory simultaneously under the GPU memory constraints. We formulate this problem as that of identifying maximal cliques in an interference graph. We approximate the interference graph using an interval graph and develop an efficient algorithm to solve the problem. Furthermore, we present two program transformations that optimize memory accesses on the GPU using the software managed scratchpad memory available in GPUs. We have prototyped the proposed compiler using the Octave system. Our experiments using this implementation show a geometric mean speedup of 12X on the GeForce 8800 GTS and 29.2X on the Tesla S1070 over baseline MATLAB execution for data parallel benchmarks. Experiments also reveal that our method provides up to 10X speedup over hand written GPUmat versions of the benchmarks. Our method also provides a speedup of 5.3X on the GeForce 8800 GTS and 13.8X on the Tesla S1070 compared to compiled MATLAB code running on the CPU.

Este trabalho tem como fulcro o desenvolvimento de um gerador de analisadores sintáticos do tipo top-down para gramáticas LL(1) com entrada gráfica da gramática, bem como uma comparação do mesmo com outros geradores em uso no mercado. Como resultado foi obtido um gerador totalmente funcional, e foi mostrado como ele é superior aos outros analisadores. São descritos detalhes da implementação e foi elaborado um manual de uso do sistema implementado em Java independente de ambientes de programação. / This thesis has as its main goal the development a parser generator using top-down syntax analysis for LL(1) grammars. Its input is a graph grammar. A comparison with available parser generators is also presented. As a result a fully executable generator, and the fact that it is superior to the other generators was demonstrated. This work contains details of the implementation, and presents a users manual of the system, which was implemented in Java. The system is independent of programming environments.

This thesis presents a technique for automatic vectorization of innermost single statement loops with a cross-iteration data dependence by analyzing data-flow to recognize frequently recurring program idioms. Recognition is carried out by matching the circular SSA data-flow found around the loop body’s φ-function against several primitive patterns, forming a tree representation of the relevant data-flow that is then pruned down to a single parameterized node, providing a high-level specification of the data-flow idiom at hand used to guide algorithmic replacement applied to the intermediate representation. The versatility of the technique is shown by presenting an implementation supporting vectorization of both a limited class of linear recurrences as well as prefix sums, where the latter shows how the technique generalizes to intermediate representations with memory state in SSA-form. Finally, a thorough performance evaluation is presented, showing the effectiveness of the vectorization technique.

This work discusses about CodAl language and Codasip framework. It describes LLVM compiling platform, LLVM IR and its possible optimizations. The result of this work is creation and implementation a proposal of global scheduling dependence on profile as extension in LLVM.

This thesis deals with the concept of the formalism of Petri nets, which allows to investigate the dynamic behavior of parallel and nondeterministic systems. Furthermore, this deals with its variant of Object-Oriented Petri Nets (OOPN), which aims to facilitate the process of modeling systems with an object-oriented approach. The aim of this master thesis is to design an internal representation of OOPN models, which is suitable for efficient interpretation and implement compiler from PNtalk language into the internal representation. Subsequently, design and implement an OOPN model interpreter using this internal representation of models, which in addition to Petri net objects, must also be able to work with a selected subset of objects from the Java language.

Tématem této práce je sémantické vyhledávání ve velkých textových datech. Cílem je navrhnout a implementovat vyhledávač, který se bude efektivně dotazovat nad sémanticky obohacenými dokumenty a prezentovat výsledky uživatelsky přívětivým způsobem. V práci jsou nejdříve analyzovány současné sémantické vyhledávače, spolu s jejich silnými a slabými stránkami. Poté je přednesen návrh nového vyhledávače s vlastním dotazovacím jazykem. Tento systém se skládá z komponent pro indexaci a dotazování se nad dokumenty, management serveru, překladače pro dotazovací jazyk a dvou klientských aplikací, webové a konzolové. Vyhledávač byl úspěšně navržen, implementován i nasazen a je veřejně dostupný na Internetu. Výsledky práce umožňují široké veřejnosti využívat sémantického vyhledávání.

This thesis describes opinions about generating code from PNtalk language into more standard object orinted languages, such as C++ or Java. Goal is to construct model of the formalism of Object Oriented Petri Nets (OOPN), represented in PNtalk languege in mentioned languages. This model should be simlpier and more effective than actual implementation in Smalltalk language. Thesis also contains ideas about abstactization the code generator to be able to generate code in various languages.

Presented is implementation of algorithm for alias analysis, which was integrated into LLVM framework. Properties and limitations of various alias analysis algorithms are discussed. Demonstrated are different approaches to working with predicates and integration of these principles with LLVM. One of the outcomes of this master's thesis is design and implementation of algorithm for profile guided if-conversion.

This paper describes methods and procedures used for code analysis and transformation. It contains basic information of a science discipline called reverse engineering and its use in information technologies. The primary objective is a construction of a generic reverse compiler or decompiler, i.e. tool that can recompile from binary form (optionally from symbolic machine code) to a high level language. This operation is highly dependent on the concrete instruction set and processor architecture. This problem is solved with description of semantic of each instruction by a special language designed for this use. The output is the high level language code and is functionally equivalent to the input. The program is therefore able to work with each instruction set and code written by it can be transformed into the chosen high level language. This proposal is implemented in practice as a part of project Lissom. Generic decompiler is completely new idea. The thesis contains entirely new techniques from theory of compilers and optimizations made by the author.