K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman Chariots problem. In ACM Transactions on Programming Languages and Systems.ACM, New York, pages 462-491, May 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....analysis of control flow diagrams. Potential for non termination Compositional interprocedural model Control flow analysis for code optimization Dependence graphs for use in code optimization Figure 2.3: Timeline of Research in Program Dependence Analysis. decades Weiser and others [23, 27, 40, 47, 56, 50] have applied program dependence analysis to a wide variety of software maintenance and understanding activities. Figure 2.3 shows a chronology of major steps in dependence analysis research. The remainder of this section contains a concise history of dependence analysis, including descriptions ....

K. Pingali and G. Bilardi. Optimal Control Dependence Computation and the Roman Chariots Problem. ACM Transactions on Programming Languages and Systems, pages 462-491, May 1997. 96

....them excellent teaching tools. key words: Dominators, Dominance Frontiers Introduction The advent of static single assignment form (ssa)hasrekindled interest in dominance and related concepts [13] New algorithms for several problems in optimization and code generation have built on dominance [8, 12, 25, 27]. In this paper, we re examine the formulation of dominance as a forward data flow problem [4, 5, 19] We present several insights that lead to a simple, general, and e#cient implementation in an iterative data flow framework. The resulting algorithm, an iterative solver that uses our ....

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

....for now to say that our construction and analysis Compare to [5, 43] See section 3 for definitions of the variables used in the complexity bounds of these two paragraphs. algorithms are efficient. All explicit control dependencies are eliminated. Some researchers (including [4] and [32]) view control dependence as a fundamental property of the CFG, and [5, 4] suggest that accurate knowledge of controldependence relations is the sole key to automatic parallelization. Often, incomplete intermediate representations are augmented with controldependence edges to express proper ....

K. Pingali and G. Bilardi. Optimal control dependence computation and the roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

....statements on program behavior. Data dependence information captures the effects of data interactions on program behavior. Tools such as program slicers use control and data dependence information for tasks such as debugging, impact analysis, and regression testing. Much research (e.g. [3, 6, 8, 21, 24, 27]) has addressed the problems of computing and utilizing intraprocedural dependences: dependences within procedures that can be computed by analyzing procedures independently. That research has considered both control and data dependence. 1 To function effectively on whole programs, however, ....

....first approach, but at the cost of some precision. 10 4. 1 Precise computation of interprocedural control dependences One way to compute interprocedural control dependences for a program P is to build the IIFG G I for P , and apply an existing algorithm, such as those described in References [3, 6, 8, 21], to G I . For non recursive programs, this approach computes precise interprocedural control dependences. In practice, this approach may be expensive. The IIFG construction inlines a procedure at each call site to that procedure; thus, the size of an IIFG may be exponential in the size of the ....

[Article contains additional citation context not shown here]

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

....statements on program behavior. Data dependence information captures the effects of data interactions on program behavior. Tools such as program slicers use control and data dependence information for tasks such as debugging, impact analysis, and regression testing. Much research (e.g. [3, 6, 8, 21, 24, 27]) has addressed the problems of computing and utilizing intraprocedural dependences: dependences within procedures that can be computed by analyzing procedures independently. That research has considered both control and data dependence. An earlier version of this paper, Computation of ....

....language 5889 unzip Zipfile extract utility 2370 4. 1 Precise computation of interprocedural control dependences One way to compute interprocedural control dependences for a program P is to build the IIFG G I for P, and apply an existing algorithm, such as those described in References [3, 6, 8, 21], to G I . For non recursive programs, this approach computes precise interprocedural control dependences. In practice, this approach may be expensive. The IIFG construction inlines a procedure at each call site to that procedure; thus, the size of an IIFG may be exponential in the size of the ....

[Article contains additional citation context not shown here]

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

....captures the effects of data interactions on program behavior. Tools such as program slicers use control and data dependence information for tasks such as debugging, impact analysis, and regression testing. Much research (e.g. Bilardi and Pingali 1996; Cytron et al. 1991; Ferrante et al. 1987; Pingali and Bilardi 1997; Pollock and Soffa 1989; Ryder and Paull 1988] has addressed the problems of computing and utilizing intraprocedural dependences: dependences within procedures that can be computed by analyzing procedures independently. That research has considered both control and data dependence. To function ....

.... computation of interprocedural control dependences One way to compute interprocedural control dependences for a program P is to build the IIFG G I for P, and apply an existing algorithm, such as those described in References [Bilardi and Pingali 1996; Cytron et al. 1991; Ferrante et al. 1987; Pingali and Bilardi 1997], to G I . For non recursive programs, this approach computes precise interprocedural control dependences. In practice, this approach may be expensive. The IIFG construction inlines a procedure at each call site to that procedure; thus, the size of an IIFG may be exponential in the size of the ....

[Article contains additional citation context not shown here]

Pingali, K. and Bilardi, G. 1997. Optimal control dependence computation and the Roman chariots problem. ACM Trans. on Programm. Lang. Syst. 19, 3 (May), 462--491.

....as a slicer, which uses control and data dependencies to identify a superset of the statements 1 that could semantically a ect another statement, can omit semantic dependencies if it utilizes inappropriate de nitions or computations of data or control dependence information. Much research (e.g. [4, 6, 8, 16, 18, 21]) has addressed the problem of computing intraprocedural dependencies: dependencies that exist within procedures and can be computed by analyzing procedures independently. To function e ectively on whole programs, however, tools that require dependence information also require information about ....

....in RN associated with n, and one labeled F from each n 2 PR to n sx ; and SX is a set of edges that connect each exit from P and each n 2 CN to n sx . Having constructed the ACFG for P , FindIntraCD( next computes the intraprocedural control dependencies for P by using an existing technique [1, 4, 6, 8, 16] (line 19) Finally, the procedure returns the set of control dependencies (lines 20 23) To illustrate the construction of an ACFG, consider procedure M from our example program. Figure 9 displays the CFG and the ACFG for M. To create the ACFG, the algorithm rst creates a super entry node, ....

[Article contains additional citation context not shown here]

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman Chariots problsm. ACM Transactions on Programming Languages and Systems, 19(3):462, May 1997.

....of Cytron et al. 7] on the SSA form, a data structure with numerous applications in program optimization. The Cytron et al. algorithm for computing the iterated dominance frontier of a set of vertices takes quadratic time in the worst case, but is competitive with other linear time algorithms [12, 16] for this problem in practice. 7 Conclusion In this paper, we have presented new, almost linear time, algorithms for two graph theoretic problems, that of constructing the dominator tree of a graph and that of computing the iterated dominance frontier of a set of vertices in a graph. Though ....

Keshav Pingali and Gianfranco Bilardi. Optimal control dependence computation and the roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

....form, a data structure that has found numerous applications in program optimization. The algorithm presented by Cytron et al. for computing the iterated dominance frontier of a set of vertices takes quadratic time in the worst case, but appears to be competitive with other linear time algorithms [13, 17] for this problem in practice. In fact, in applications such as constructing the SSA form, which require computing the iterated dominance frontier of a number of different sets of vertices, the Cytron et al. algorithm has been shown to perform better than the Sreedhar et al. linear time algorithm ....

....we have presented new, almost linear time, algorithms for two graph theoretic problems, that of constructing the dominator tree of a graph and that of computing the iterated dominance frontier of a set of vertices in a graph. Though linear time algorithms are already known for these problems [1, 4, 13, 17], we believe that these new algorithms are interesting because of the approach we take In particular, we have utilized the above two applications as a vehicle for understanding the concepts of loops and loop nesting forests in graphs. We have shown how three previously defined loop nesting ....

Keshav Pingali and Gianfranco Bilardi. Optimal control dependence computation and the roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

....(LT) algorithm is extremely efficient in practice. Reducing the asymptotic time complexity of finding dominators to O(n m) is an interesting theoretical exercise. Furthermore, various results in compiler theory rely on the existence of a linear time dominators algorithm; Pingali and Bilardi [19] give an example and further references. Harel [12] claimed a linear time dominators algorithm, but careful examination of his abstract reveals problems with his arguments. Alstrup et al. 3] detail some of the problems with Harel s approach and offer a linear time algorithm that employs powerful, ....

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--555, 1997. 27

....Assignment form, a data structure that has found numerous applications in program optimization. The algorithm presented by Cytron et al. for computing the iterated dominance frontier of a set of vertices takes quadratic time in the worst case, but is competitive with other linear time algorithms [16, 17, 12, 18] for this problem in practice. In fact, in applications such as constructing the SSA form, which require computing the iterated dominance frontier of a number of different sets of vertices, the Cytron et al. algorithm has been shown to perform better than the Sreedhar et al. linear time algorithm ....

....we have presented new, almost linear time, algorithms for two graph theoretic problems, that of constructing the dominator tree of a graph and that of computing the iterated dominance frontier of a set of vertices in a graph. Though linear time algorithms are already known for these problems [1, 4, 12, 18], we believe that these new algorithms are interesting because of the approach we take In particular, we have utilized the above two applications as a vehicle for understanding the concepts of loops and loop nesting forests in graphs. We have shown how three previously defined loop nesting ....

Keshav Pingali and Gianfranco Bilardi. Optimal control dependence computation and the roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

....as a slicer, which uses control and data dependencies to identify a superset of the statements that could semantically affect another statement, can omit semantic dependencies if it utilizes inappropriate definitions or computations of data or control dependence information. Much research (e.g. [4, 6, 8, 17, 19, 21]) has addressed the problem of computing intraprocedural dependencies: dependencies that exist within procedures and can be computed by analyzing procedures independently. To function effectively on whole programs, however, tools that require dependence information also require information about ....

....to the node in RN associated with n, and one labeled F from each n 2 PR to nsx ; and SX is a set of edges that connect each exit from P and each n 2 CN to nsx . Given the ACFG G A for P i , ComputeInterCD computes the intraprocedural control dependencies for P i using an existing technique [1, 4, 6, 8, 17]. To illustrate the construction of an ACFG, consider procedure M from our example program. Figure 4 displays the CFG and the ACFG for M. To create the ACFG, the algorithm first replaces node 5 with call node 5a and return node 5b, and replaces node 6 with call node 6a and return node 6b. Next, ....

[Article contains additional citation context not shown here]

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman Chariots problsm. ACM Transactions on Programming Languages and Systems, 19(3):462, May 1997.

....previous dominators algorithms for flowgraph sizes that appear in practice. Reducing the asymptotic time complexity of finding dominators to O(n m) is an interesting theoretical exercise. Furthermore, various results in compiler theory rely on the existence of a linear time dominators algorithm; Pingali and Bilardi [1997] give an example and further references. Harel [1985] claimed a linear time dominators algorithm, but careful examination of his abstract reveals problems with his arguments. Alstrup et al. 1997] detail some of the problems with Harel s approach and offer a linear time algorithm that employs ....

Pingali, K. and Bilardi, G. 1997. Optimal control dependence computation and the Roman chariots problem. ACM Trans. Program. Lang. Syst. 19, 3, 462--555.

....Notes inComputer Science, pages 1 15. the Quick Propagation Graph (QPG) and present a linear time algorithm for constructing it. In general, the Quick Propagation Graph is not as compact as the Sparse Evaluation Graph. Cytron and Ferrante [4] Sreedhar and Gao [16] and Pingali and Bilardi [14, 15] improve upon the efficiency of the original Choi et al. algorithm for constructing the Sparse Evaluation Graph. Duesterwald et al. 8] show how a congruence partitioning technique can be used to construct an equivalent flow graph, which we believe is the same as the standard Sparse Evaluation ....

....form [5, 6] which was used to solve various analysis problems, such as constant propagation and redundancy elimination, more efficiently. Choi et al. 2] generalized the idea and defined the Sparse Evaluation Graph. Cytron and Ferrante [4] Sreedhar and Gao [16] and Pingali and Bilardi [14, 15] improve upon the efficiency of the original Choi et al. algorithm for constructing the Sparse Evaluation Graph. We will discuss the relative efficiencies of the various algorithms in detail soon. Johnson et al. 11, 10] define a different equivalent flow graph called the Quick Propagation ....

[Article contains additional citation context not shown here]

Keshav Pingali and Gianfranco Bilardi. Optimal control dependence computation and the roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

No context found.

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman Chariots problem. In ACM Transactions on Programming Languages and Systems.ACM, New York, pages 462-491, May 1997.

....when several sets M(S 1 ) M(S 2 ) have to be computed, since the DF information is rederived from scratch for each query. Both issues are addressed in Section 6, with the introduction of the augmented dominator tree (ADT ) a data structure similar to the augmented postdominator tree [PB97] The first issue is addressed by constructing the DF graph lazily as needed by the reachability computation. The idea of lazy algorithms is quite general and involves computing f(g(x) by computing only that portion of g(x) that is required to produce the output of f [Haskell] The second issue ....

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman Chariots problem. In ACM Transactions on Programming Languages and Systems. ACM, New York, pages 462-491, May 1997.

....when several sets J(S 1 ) J(S 2 ) have to be computed, since the DF information is rederived from scratch for each query. Both issues are addressed in Section 6, with the introduction of the augmented dominator tree (ADT ) a data structure similar to the augmented postdominator tree [PB97] The first issue is addressed by constructing the DF graph lazily as needed by the reachability computation. The idea of lazy algorithms is quite general and involves computing f(g(x) by computing only that portion of g(x) that is required to produce the output of f [Haskell] The second issue ....

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman Chariots problem. In ACM Transactions on Programming Languages and Systems. ACM, New York, pages 462-491, May 1997.

No context found.

Keshav Pingali and Gianfranco Bilardi. Optimal control dependence computation and the Roman Chariots Problem. ACM Transactions on Programming Languages and Systems, 19(3):462-491, May 1997.

No context found.

K. Pingali and G. Bilardi. Optimal control dependence computation and the Roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462--491, May 1997.

No context found.

K. Pingali and G. Bilardi. Optimal control dependence computation and the roman chariots problem. ACM Transactions on Programming Languages and Systems, 19(3):462-491, May 1997.

No context found.

K. Pingali and G. Bilardi. Optimal Control Dependence Computation and the Roman Chariots Problem. ACM Transactions on Programming Languages and Systems, pages 462-491, May 1997. 96

No context found.

Keshav Pingali & Gianfranco Bilardi, "Optimal Control Dependence Computation and the Roman Chariots Problem," Trans. Programming Lang. and Syst. (May, 1997), #http://www.cs.cornell.edu/Info/ Projects/Bernoulli/papers/toplas97.ps#.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC