L’approche synchrone aux systèmes réactifs, où le temps global est une séquence d’instants discrets, a été proposée afin de faciliter la conception des systèmes embarqués critiques. Des systèmes synchrones sont souvent réalisés sur des architectures asynchrones pour des raisons de performance ou de contraintes physiques de l’application. Une répartition d’un système synchrone sur une architecture asynchrone nécessite des protocoles de communication et de synchronisation pour préserver la sémantique synchrone. En pratique, les protocoles peut avoir un coût important qui peut entrer en conflit avec les contraintes de l’application comme, par exemple, la taille de mémoire disponible, le temps de réaction, ou le débit global.L’approche quasi-synchrone utilise des composants synchrones avec des horloges indépendantes. Les composants communiquent par échantillonnage de mémoire partagée ou par des tampons FIFO. On peut exécuter un tel système de façon synchrone, où toutes les horloges avancent simultanément, ou de façon asynchrone avec moins de contraintes sur les horloges, sans ajouter des protocoles .Plus les contraintes sont relâchées, plus de comportements se rajoutent en fonction de l’entrelacement des tics des horloges. Dans le cas de systèmes flots de données, un comportement est différent d’un autre si les valeurs ou le cadencement ont changé. Pour certaines classes de systèmes l’occurrence des déviations est acceptable, tant que la fréquence de ces événements reste bornée.Nous considérons des limites dures sur la fréquence des déviations avec ce que nous appelons les exigences faiblement dures, par exemple, le nombre maximal d’éléments divergents d’un flot par un nombre d’éléments consécutifs.Nous introduisons des limites de dérive sur les apparitions relatives des paires d’événements récurrents comme les tics d’une horloge, l’occurrence d’une différence,ou l’arrivée d’un message. Les limites de dérive expriment des contraintes entre les horloges, par exemple, une borne supérieure de deux tics d’une horloge entre trois tics consécutifs d’une autre horloge. Les limites permettent également de caractériser les exigences faiblement dures. Cette thèse présente des analyses pour la vérification et l’inférence des exigences faiblement dures pour des programmes de flots de données synchrones étendu avec de la communication asynchrone par l’échantillonnage de mémoire partagée où les horloges sont décrites par des limites de dérive. Nous proposons aussi une analyse de performance des systèmes répartis avec de la communication par tampons FIFO, en utilisant les limites de dérive comme abstraction. / The synchronous approach to reactive systems, where time evolves by globally synchronized discrete steps, has proven successful for the design of safetycriticalembedded systems. Synchronous systems are often distributed overasynchronous architectures for reasons of performance or physical constraintsof the application. Such distributions typically require communication and synchronizationprotocols to preserve the synchronous semantics. In practice, protocolsoften have a significant overhead that may conflict with design constraintssuch as maximum available buffer space, minimum reaction time, and robustness.The quasi-synchronous approach considers independently clocked, synchronouscomponents that interact via communication-by-sampling or FIFO channels. Insuch systems we can move from total synchrony, where all clocks tick simultaneously,to global asynchrony by relaxing constraints on the clocks and withoutadditional protocols. Relaxing the constraints adds different behaviors dependingon the interleavings of clock ticks. In the case of data-flow systems, onebehavior is different from another when the values and timing of items in a flowof one behavior differ from the values and timing of items in the same flow ofthe other behavior. In many systems, such as distributed control systems, theoccasional difference is acceptable as long as the frequency of such differencesis bounded. We suppose hard bounds on the frequency of deviating items in aflow with, what we call, weakly-hard requirements, e.g., the maximum numberdeviations out of a given number of consecutive items.We define relative drift bounds on pairs of recurring events such as clockticks, the occurrence of a difference or the arrival of a message. Drift boundsexpress constraints on the stability of clocks, e.g., at least two ticks of one perthree consecutive ticks of the other. Drift bounds also describe weakly-hardrequirements. This thesis presents analyses to verify weakly-hard requirementsand infer weakly-hard properties of basic synchronous data-flow programs withasynchronous communication-by-sampling when executed with clocks describedby drift bounds. Moreover, we use drift bounds as an abstraction in a performanceanalysis of stream processing systems based on FIFO-channels.

A number of questions regarding programs involving heap-based data structures can be phrased as questions about numeric properties of those structures. A data structure traversal might terminate if the length of some path is eventually zero or a function to remove n elements from a collection may only be safe if the collection has size at least n. In this thesis, we develop proof methods for reasoning about the connection between heap-manipulating programs and numeric programs. In addition, we develop an automatic method for producing numeric abstractions of heap-manipulating programs. These numeric abstractions are expressed as simple imperative programs over integer variables and have the feature that if a property holds of the numeric program, then it also holds of the original, heap-manipulating program. This is true for both safety and liveness. The abstraction procedure makes use of a shape analysis based on separation logic and has support for user-defined inductive data structures. We also discuss a number of applications of this technique. Numeric abstractions, once obtained, can be analyzed with a variety of existing verification tools. Termination provers can be used to reason about termination of the numeric abstraction, and thus termination of the original program. Safety checkers can be used to reason about assertion safety. And bound inference tools can be used to obtain bounds on the values of program variables. With small changes to the program source, bounds analysis also allows the computation of symbolic bounds on memory use and computational complexity.

Made available in DSpace on 2014-06-11T19:30:32Z (GMT). No. of bitstreams: 0 Previous issue date: 2009-10-09Bitstream added on 2014-06-13T18:40:52Z : No. of bitstreams: 1 boccardo_dr_dr_ilha.pdf: 1178776 bytes, checksum: cdd885f0beff962757e3b9de59ce0832 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Ofusca c~ao de c odigo tem por nalidade di cultar a detec c~ao de propriedades intr nsecas de um algoritmo atrav es de altera c~oes em sua sintaxe, entretanto preservando sua sem antica. Desenvolvedores de software usam ofusca c~ao de c odigo para defender seus programas contra ataques de propriedade intelectual e para aumentar a seguran ca do c odigo. Por outro lado, programadores maliciosos geralmente ofuscam seus c odigos para esconder comportamento malicioso e para evitar detec c~ao pelos anti-v rus. Nesta tese, e introduzido um m etodo para realizar an alise com sensitividade ao contexto em bin arios com ofuscamento de chamada e retorno de procedimento. Para obter sem antica equivalente, estes bin arios utilizam opera c~oes diretamente na pilha ao inv es de instru c~oes convencionais de chamada e retorno de procedimento. No estado da arte atual, a de ni c~ao de sensitividade ao contexto est a associada com opera c~oes de chamada e retorno de procedimento, assim, an alises interprocedurais cl assicas n~ao s~ao con aveis para analisar bin arios cujas opera c~oes n~ao podem ser determinadas. Uma nova de ni c~ao de sensitividade ao contexto e introduzida, baseada no estado da pilha em qualquer instru c~ao. Enquanto mudan cas em contextos a chamada de procedimento s~ao intrinsicamente relacionadas com transfer encia de controle, assim, podendo ser obtidas em termos de caminhos em um grafo de controle de uxo interprocedural, o mesmo n~ao e aplic avel para mudan cas em contextos a pilha. Um framework baseado em interpreta c~ao abstrata e desenvolvido para avaliar contexto baseado no estado da pilha e para derivar m etodos baseado em contextos a chamada de procedimento para uso com contextos baseado no estado da pilha. O metodo proposto n~ao requer o uso expl cito de instru c~oes de chamada e retorno de procedimento, por em depende do... / A code obfuscation intends to confuse a program in order to make it more di cult to understand while preserving its functionality. Programs may be obfuscated to protect intellectual property and to increase security of code. Programs may also be obfuscated to hide malicious behavior and to evade detection by anti-virus scanners. We introduce a method for context-sensitive analysis of binaries that may have obfuscated procedure call and return operations. These binaries may use direct stack operators instead of the native call and ret instructions to achieve equivalent behavior. Since de nition of context-sensitivity and algorithms for context-sensitive analysis has thus far been based on the speci c semantics associated to procedure call and return operations, classic interprocedural analyses cannot be used reliably for analyzing programs in which these operations cannot be discerned. A new notion of context-sensitivity is introduced that is based on the state of the stack at any instruction. While changes in calling-context are associated with transfer of control, and hence can be reasoned in terms of paths in an interprocedural control ow graph (ICFG), the same is not true for changes in stackcontext. An abstract interpretation based framework is developed to reason about stackcontext and to derive analogues of call-strings based methods for the context-sensitive analysis using stack-context. This analysis requires the knowledge of how the stack, rather the stack pointer, is represented and on the knowledge of operators that manipulate the stack pointer. The method presented is used to create a context-sensitive version of Venable et al.'s algorithm for detecting obfuscated calls. Experimental results show that the contextsensitive version of the algorithm generates more precise results and is also computationally more e cient than its context-insensitive counterpart.

This dissertation shows that the bounded property verification of hardware Register Transfer Level (RTL) designs can be efficiently performed by precise abstract interpretation of a software representation of the RTL. The first part of this dissertation presents a novel framework for RTL verification using native software analyzers. To this end, we first present a translation of the hardware circuit expressed in Verilog RTL into the software in C called the software netlist. We then present the application of native software analyzers based on SAT/SMT-based decision procedures as well as abstraction-based techniques such as abstract interpretation for the formal verification of the software netlist design generated from the hardware RTL. In particular, we show that the path-based symbolic execution techniques, commonly used for automatic test case generation in system softwares, are also effective for proving bounded safety as well as detecting bugs in the software netlist designs. Furthermore, by means of experiments, we show that abstract interpretation techniques, commonly used for static program analysis, can also be used for bounded as well as unbounded safety property verification of the software netlist designs. However, the analysis using abstract interpretation shows high degree of imprecision on our benchmarks which is handled by manually guiding the analysis with various trace partitioning directives. The second part of this dissertation presents a new theoretical framework and a practical instantiation for automatically refining the precision of abstract interpretation using Conflict Driven Clause Learning (CDCL)-style analysis. The theoretical contribution is the abstract interpretation framework that generalizes CDCL to precise safety verification for automatic transformer refinement called Abstract Conflict Driven Learning for Safety (ACDLS). The practical contribution instantiates ACDLS over a template polyhedra abstract domain for bounded safety verification of the software netlist designs. We experimentally show that ACDLS is more efficient than a SAT-based analysis as well as sufficiently more precise than a commercial abstract interpreter.

We have written an optimising compiler for a typeless, imperative, modular programming language. The optimiser, which works on a 3-address intermediate representation generated from the source program, uses some novel techniques described in this thesis. The techniques are universally applicable, although some are particularly useful in typeless compilation. We present a new register allocation and assignment scheme. Unlike traditional "colouring" allocators, our method separates the problem into distinct allocation and assignment phases. The former is achieved by using an iterative process to extend a local (within basic blocks) allocation method to the global (across basic blocks) domain. This obviates the need for a sophisticated assignment algorithm; we show how to use simple heuristics to assign registers after allocation. We also present a simple method for identifying loops in a program's intermediate representation and assigning loop nesting levels. Unlike traditional methods, this does not rely on the concept of flowgraph dominators, and is able to deal sensibly with irreducible flowgraphs and "unstructured" loops that interlock or partially overlap. The major part of the thesis concerns value range analysis. Based on the theoretical framework of abstract interpretation, we describe an analysis of the intermediate code that predicts safe approximations to the run-time value ranges of variables and memory used by the program being compiled. To be useful in compiling a typeless language, this analysis must be able to handle values of different kinds (integers, pointers, function addresses, etc.) We show how we can subsume some traditional optimisation techniques, such as constant propagation, into more powerful methods that take advantage of value range information to optimise a wider variety of cases. We also show how this information can be used to recover most of the benefits of types, without sacrificing the flexibility of typelessness. Besides the above, value range analysis allows a number of optimisations that were heretofore impossible. Many of these are improvements to register allocation; we investigate better treatments for variables that can be accessed by address. We also describe a method of removing memory accesses by allowing variables that are simultaneously live to share registers, and suggest a similar scheme for values stored in memory. Finally, we show how the results of value range analysis can be shared across different program modules and different compiler runs. The method used is powerful enough to be useful, but simple enough to integrate with old code that cannot be recompiled. Inter-modular optimisation can be transparent to the user, improving the results of value range analysis within a module without altering its functionality; or it can be visible, optimising modules with respect to each other.

L’analyse statique de programmes est un domaine crucial en compilation, en optimisation, et en validation de logiciels. Les structures de données complexes (tableaux, listes, graphes...), omniprésentes dans les programmes, posent des problèmes difficiles, du fait qu’elles représentent des ensembles de données de taille importante ou inconnue, et que l’adressage des données dans ces ensembles est calculé (indexation, indirection). La plupart des travaux sur l’analyse des structures de données concernent la vérification de la correction des accès aux données (vérification que les indices d’un tableau sont dans les bornes, que les pointeurs ne sont pas nuls, “shape analysis”). L’analyse du contenu des structures de données est encore peu abordée. A Verimag, ce domaine a été abordé récemment, et a donné lieu à de premiers résultats sur l’analyse de tableaux unidimensionnels. Une méthode d’analyse de programmes simples a été proposée [1], qui permet de découvrir des propriétés des contenus de tableaux, comme par exemple que le résultat d’un programme de tri est bien un tableau trié. Un autre type de propriétés, dites “non positionnelles” a aussi été considéré [2], qui concerne le contenu global d’un tableau, indépendamment de son rangement: par exemple, on montre que le résultat d’un tri est une permutation du tableau initial. Ces premiers résultats sont très encourageants, mais encore embryonnaires. L’objectif du travail de thèse proposé est de les étendre dans plusieurs directions. Notre analyse de propriétés positionnelles est capable de découvrir des relations point- à-point entre des “tranches” de tableaux (ensembles de cellules consécutives). Les extensions envisagées concernent les tableaux multidimensionnels, les ensembles de cellules non nécessairement consécutives, et les structures de données plus générales. Concernant les propriétés non positionnelles, les premiers résultats sont limités aux égalités de contenus de tableaux. Ils doivent être étendus à des relations plus complexes (inclusions, sommes disjointes...) et à d’autres structures de données. Ce travail prend place dans le projet ASOPT (“Analyse statique et optimisation”), accepté dans le programme Arpège de l’ANR en 2008. Références : [1] N. Halbwachs, M. Péron. Discovering properties about arrays in simple programs. ACM Conference on Programming Language Design and Implementation, PLDI 2008. Tucson (Az.), juin 2008. [2] V. Perrelle. Analyse statique du contenu de tableaux, propriétés non positionnelles. Rapport de M2R, Master Parisien de Recherche en Informatique, septembre 2008. / Static analysis is key area in compilation, optimization and software validation. The complex data structures (arrays, dynamic lists, graphs...) are ubiquitous in programs, and can be challenging, because they can be large or of unbounded size and accesses are computed. (through indexing or indirections). Whereas the verification of the validity of the array accesses was one of the initial motivations of abstract interpretation, the discovery of properties about array contents was only adressed recently. Most of the analyses of array contents are based on a partitioning of the arrays. Then, they try to discover properties about each fragment of this partition. The choice of this partition is a difficult problem and each method have its flaw. Moreover, classical representations of array partitions induce an exponential complexity for these analyzes. In this thesis, we generalize the concept of array partitioning into the concept of "fragmentation" which allow overlapping fragments, handling potentially empty fragments and selecting specialized relations. On the other hand, we propose an abstraction of these fragmentations in terms of graphs called "slices diagrams" as well as the operations to manipulate them and ensuring a polynomial complexity. Finally, we propose a new criterion to compute a semantic fragmentation inspired by the existing ones which attempt to correct their flaws. These methods have been implemented in a static analyzer. Experimentations shows that the analyzer can efficiently and precisly prove some challenging exemples in the field of static analysis of programs manipulating arrays.

La rapidité, la consommation énergétique et l'efficacité des systèmes logiciels et matériels sont devenues les préoccupations majeures de la communauté informatique de nos jours. Gérer de manière correcte et efficace les problématiques mémoire est essentiel pour le développement des programmes de grande tailles sur des architectures de plus en plus complexes. Dans ce contexte, cette thèse contribue aux domaines de l'analyse mémoire et de la compilation tant sur les aspects théoriques que sur les aspects pratiques et expérimentaux. Outre l'étude approfondie de l'état de l'art des analyses mémoire et des différentes limitations qu'elles montrent, notre contribution réside dans la conception et l'évaluation de nouvelles analyses qui remédient au manque de précision des techniques publiées et implémentées. Nous nous sommes principalement attachés à améliorer l'analyse de pointeurs appartenant à une même structure de données, afin de lever une des limitations majeures des compilateurs actuels. Nous développons nos analyses dans le cadre général de l'interprétation abstraite « non dense ». Ce choix est motivé par les aspects de correction et d'efficacité : deux critères requis pour une intégration facile dans un compilateur. La première analyse que nous concevons est basée sur l'analyse d'intervalles des variables entières ; elle utilise le fait que deux pointeurs définis à l'aide d'un même pointeur de base n'aliasent pas si les valeurs possibles des décalages sont disjointes. La seconde analyse que nous développons est inspirée du domaine abstrait des Pentagones ; elle génère des relations d'ordre strict entre des paires de pointeurs comparables. Enfin, nous combinons et enrichissons les deux analyses précédentes dans un cadre plus général. Ces analyses ont été implémentées dans le compilateur LLVM. Nous expérimentons et évaluons leurs performances, et les comparons aux implémentations disponibles selon deux métriques : le nombre de paires de pointeurs pour lesquelles nous inférons le non-aliasing et les optimisations rendues possibles par nos analyses / This thesis was motivated by the emergence of massively parallel processing and supercomputingthat tend to make computer programming extremely performing. Speedup, the power consump-tion, and the efficiency of both software and hardware are nowadays the main concerns of theinformation systems community. Handling memory in a correct and efficient way is a step towardless complex and more performing programs and architectures. This thesis falls into this contextand contributes to memory analysis and compilation fields in both theoretical and experimentalaspects.Besides the deep study of the current state-of-the-art of memory analyses and their limitations,our theoretical results stand in designing new algorithms to recover part of the imprecisionthat published techniques still show. Among the present limitations, we focus our research onthe pointer arithmetic to disambiguate pointers within the same data structure. We develop ouranalyses in the abstract interpretation framework. The key idea behind this choice is correctness,and scalability: two requisite criteria for analyses to be embedded to the compiler construction.The first alias analysis we design is based on the range lattice of integer variables. Given a pair ofpointers defined from a common base pointer, they are disjoint if their offsets cannot have valuesthat intersect at runtime. The second pointer analysis we develop is inspired from the Pentagonabstract domain. We conclude that two pointers do not alias whenever we are able to build astrict relation between them, valid at program points where the two variables are simultaneouslyalive. In a third algorithm we design, we combine both the first and second analysis, and enhancethem with a coarse grained but efficient analysis to deal with non related pointers.We implement these analyses on top of the LLVM compiler. We experiment and evaluate theirperformance based on two metrics: the number of disambiguated pairs of pointers compared tocommon analyses of the compiler, and the optimizations further enabled thanks to the extraprecision they introduce

In a real-time system, it is crucial to ensure that all tasks of the system holdtheir deadlines. A missed deadline in a real-time system means that the systemhas not been able to function correctly. If the system is safety critical, this canlead to disaster. To ensure that all tasks keep their deadlines, the Worst-CaseExecution Time (WCET) of these tasks has to be known. This can be done bymeasuring the execution times of a task, however, this is inflexible, time consumingand in general not safe (i.e., the worst-casemight not be found). Unlessthe task is measured with all possible input combinations and configurations,which is in most cases out of the question, there is no way to guarantee that thelongest measured time actually corresponds to the real worst case.Static analysis analyses a safe model of the hardware together with thesource or object code of a program to derive an estimate of theWCET. This estimateis guaranteed to be equal to or greater than the real WCET. This is doneby making calculations which in all steps make sure that the time is exactlyor conservatively estimated. In many cases, however, the execution time of atask or a program is highly dependent on the given input. Thus, the estimatedworst case may correspond to some input or configuration which is rarely (ornever) used in practice. For such systems, where execution time is highly inputdependent, a more accurate timing analysis which take input into considerationis desired.In this thesis we present a framework based on abstract interpretation andcounting of possible semantic states of a program. This is a general methodof WCET analysis, which is language independent and platform independent.The two main applications of this framework are a loop bound analysis and aparametric analysis. The loop bound analysis can be used to quickly find upperbounds for loops in a program while the parametric framework provides aninput-dependent estimation of theWCET. The input-dependent estimation cangive much more accurate estimates if the input is known at run-time. / PROGRESS

Using contracts such as method preconditions, postconditions, and class invariants in code is an approach to preventing bugs in object-oriented programs. In the .NET Framework, this is possible due to the framework of Code Contracts, which includes Clousot, a tool for static program analysis based on abstract interpretation. Although string is a fundamental type in .NET programs, Clousot does not have a usable support for analysis of string values. In this thesis, we explore the specifics of string manipulation in the C# language and in the .NET Framework, and show how they can be covered by static analysis. Our approach is to use the methods of the String class and a subset of regular expressions to specify string properties in code, and to use abstract interpretation with non-relational abstract domains to reason about those properties. We chose a small number of already published abstract domains for strings, which vary in their complexity and ability to represent different properties. We adapted those domains to our setting, including definitions of abstract semantics for the supported string methods. We implemented the abstract domains in Clousot in a way that cooperates with numerical analysis and allows adding more string abstract domains in the future. Powered by TCPDF (www.tcpdf.org)

Cette thèse propose un nouveau cadre pour la composition de domaines abstraits. L'idée principale en est l'organisation d'une sémantique abstraite suivant la distinction usuelle entre expressions et instructions, en cours dans la plupart des langages impératifs. La définition d'une sémantique abstraite peut alors se diviser entre abstractions de valeurs et abstractions d'états (ou domaine abstrait). Les abstractions de valeurs représentent les valeurs possibles d'une expression en un point donné, et assurent l'interprétation de la sémantique des expressions. Les abstractions d'états représentent les états machines qui peuvent se produire lors de l'exécution d'un programme, et permettent d'interpréter la sémantique des instructions. De ce choix de conception découle naturellement un élégant système de communication entre abstractions. Lors de l'interprétation d'une instruction, les abstractions d'états peuvent échanger des informations au moyen d'abstractions de valeurs, qui expriment des propriétés à propos des expressions. Les valeurs forment donc une interface de communication entre états abstraits, mais sont également des éléments canoniques de l'interprétation abstraite. Ils peuvent donc eux-même être combinés par les moyens existants de composition d'abstractions, permettant encore davantage d'interactions entre les composants des sémantiques abstraites. Cette thèse explore les possibilités offertes par cette nouvelle architecture des sémantiques abstraites. Nous décrivons en particulier des stratégies efficaces pour le calcul d'abstractions de valeurs précises à partir des propriétés inférées par les domaines, et nous illustrons les différentes possibilités d'interactions que ce système offre. L'architecture que nous proposons inclue également une collaboration directe des abstractions pour l'émission des alarmes qui signalent les erreurs possibles du programme analysé. Nous proposons également un mécanisme permettant d'interagir avec les composants d'une combinaison générique de types OCaml. Nous utilisons des GADT pour encoder la structure interne d'une combinaison, et construisons automatiquement les fonctions d'injection et de projection entre le produit et ses composants. Cette fonctionnalité permet d'établir une communication directe entre les différentes abstractions d'un interpréteur abstrait. Enfin, une dernière contribution de cette thèse est l'extension automatique de domaines abstraits à l'aide de prédicats logiques qui évitent les pertes d'information aux points de jonction. De fait, lorsque plusieurs chemins d'exécution se rejoignent, un domaine abstrait doit représenter les comportements possibles de chacun des chemins, ce qui engendre souvent des pertes de précision. Pour remédier à cette limitation, nous proposons de propager un ensemble d'états abstraits, munis chacun d'un prédicat qui indique sous quelle condition l'état est valable. Contrairement à d'autres approches, notre analyse ne maintient pas une stricte partition des états abstraits, car les prédicats utilisés ne sont pas mutuellement exclusifs. Cette particularité rend possible des optimisations cruciales pour le passage à l'échelle de cette technique, confirmée par nos résultats expérimentaux sur un programme industriel généré. L'ensemble du système de composition des abstractions proposé dans cette thèse a été mis en œuvre dans EVA, la nouvelle version de l'interpréteur abstrait de Frama-C. EVA a été spécifiquement conçu pour faciliter l'introduction de nouvelles abstractions et permettre des interactions riches entre ces abstractions. Grâce à son architecture modulaire et extensible, cinq nouveaux domaines abstraits ont pu être introduit dans l'analyseur en moins d'un an, améliorant ainsi tant ses capacités que sa précision. / This thesis proposes a new framework for the combination of multiple domains in the abstract interpretation theory. Its core concept is the structuring of the abstract semantics by following the usual distinction between expressions and statements. This can be achieved by a convenient architecture where abstractions are separated in two layers: value abstractions, in charge of the expression semantics, and state abstractions —or abstract domains—, in charge of the statement semantics. This design leads naturally to an elegant communication system where the abstract domains, when interpreting a statement, interact and exchange information through value abstractions, that express properties about expressions. While the values form the communication interface between domains, they are also standard elements of the abstract interpretation framework. The communication system is thus embedded in the abstract semantics, and the usual tools of abstract interpretation apply naturally to value abstractions. For instance, different kinds of value abstractions can be composed through the existing methods of combination of abstractions, enabling even further interaction between the components of the abstract semantics. This thesis explores the possibilities offered by this framework. We discuss efficient strategies to compute precise value abstractions from the information inferred by abstract domains, and illustrate the means of communication between different state abstractions. Our architecture also features a direct collaboration for the emission of alarms that report the possible errors of a program. We also proposes a mechanism to enable interacting with the components of a modular combination of OCaml types. We use GADT to encode the inner shape of a combination, and automatically build injection and projection functions between a product of datatypes and its components. This allows direct communications between the abstractions of an abstract interpreter. Finally, a last contribution of this thesis is the automatic extension of abstract domains to track sets of disjunctive abstract states, each one being qualified with a predicate for which the state holds. This enhances the precision of an abstract semantics at join points, when several possible paths of a program execution meet. At these points, predicates preserve the information usually lost by the merge of abstract states. Unlike other approaches, the analysis does not maintain a strict partition of the abstract states, as the predicates we use are not mutually exclusive. This design enables some optimizations that are crucial for scalability, as confirmed by our experimental results on an industrial, generated program. The general system of abstractions combination has been implemented within EVA, the new version of the abstract interpreter provided by the Frama-C platform. Thus, Eva enjoys a modular and extensible architecture designed to facilitate the introduction of new abstractions and to enable rich interactions between them. Thanks to this work, five new domains from the literature have been implemented in less than a year, enhancing the scope and the precision of the analyzer.