D. Whitfield and M. Soffa, "An approach to ordering optimizing transformations, " ACM Symposium on Principles and Practice of Parallel Programming, Vol. 25, No. 3, pp. 137-146, 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....prerequisites in the engines. Is it possible to avoid reprogramming of engines and embed them transparently into such a context Another parameter that influences the optimization results is how the optimization engines are ordered. Because certain optimizations enable or disable others (see [WS90]) the optimizer writer should be given a means to embed an engine into several ordering contexts, e.g. into loops that run until no optimization can be applied anymore (exhaustive transformation) or into loops whose iteration number is configured by the machine. Of course exhaustive ....

....Embedding into alterable engine orderings In this section we leave the field of register allocation and turn to complex optimization orderings. We show how a simple linear optimization engine can be reconfigured into a complex one with a sophisticated ordering. This is exemplified by the proposal [WS90] which takes several dependencies between the used optimizations into account. First we show how a simple optimizer may be specified in EDL just by concatenating some simple engines. For simplicity we ignore the side effect description here. invariant code motion propagation constant dead ....

[Article contains additional citation context not shown here]

D. Whitfield and M. L. Soffa. An approach to ordering optimizing transformations. In ACM Conference on Principles and Practice of Parallel Programming (PPOPP), 1990. This article was processed using the L a T E X macro package with LLNCS style

....than repeated sequences of the two analyses run separately. Cousot and Cousot [11] also point out that such interactions can arise. However, in all these cases, the composition needs to be done manually by defining special flow functions over the combined dataflow information. Whitfield and So#a [31, 32] have developed a framework for examining the interactions between di#erent optimizations. By analyzing the pre and post conditions of optimizations, their framework can determine if one optimization helps or hinders another optimization. This information can then be used to select an order in ....

Debbie Whitfield and Mary Lou So#a. An approach to ordering optimizing transformations. In Second ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (2nd PPOPP'90), SIGPLAN Notices, pages 137--146, March 1990.

....in reducing the search space. A transformation is enabling, if its application does not improve objective function, but makes feasible application of another transformation which significantly improve the quality of design. Enabling and disabling effects among transformations is summarized in [Whi90, Pot91b] 3 Asymptotic analysis information. Often it is of interest to analyze the effectiveness of transformations as the size of target problem instance increases. This analysis is not of just theoretical interest because there is a simple way to increase the size of any ASIC computation on infinite ....

D. Whitfield, M. L. Sofia: "An Approach to Ordering Optimizing Transformations", Second ACM SIGPLAN .qyrnpo.qium on Principles and Practice of Parallel Programming, pp. 137-147, 1990. 3337

....conceivable, however, that there exists a third transformation TC that enables TB again, so that we can have the sequence of steps TA # TC # TB . It is not obvious whether our methods can be extended to detect this situation. In their papers on exploring code improving transformations [8, 10], Whitfield and Soffa note that due to the lack of a commonly accepted formal notation for compiler optimisations, it is hard to analyse a set of such optimisations. They proposed the Genesis language as a basis for such a study, and they showed how it enabled an informal but rigorous analysis of ....

D. L. Whitfield and M. L. Soffa. An approach to ordering optimizing transformations. In ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, pages 137-- 146, 1990.

....used to guide transformation decisions. 1 Introduction We consider the problem of automatically guiding program transformations despite incomplete information. Guidance requires an infrastructure that supports queries of the form, under what circumstances should I apply this transformation [32, 27, 2, 16, 30]. Answering these queries in the face of complicated program structures, unknown target architecture, and lack of knowledge of the input data requires a combined compile time runtime solution. In this paper, we present our solution for automatically guiding locality transformations: the modal ....

D. Whitfield and M. L. So#a. An approach to ordering optimizing transformations. In Principles and Practice of Parallel Programming, pages 137--146, Seattle, WA, Mar. 1990.

....moves can be performed very fast. But transformations are dependent, and their applicability always has to be checked prior to execution, which makes them relatively slow. The ordering problem can be tackled by formalising the preconditions and postconditions of a limited set of transformations [15]. From this, the enabling and disabling relationships between transformations are derived, and represented in a graph. An ordering is obtained from levelling the graph. Cycles in the graph are resolved by allowing the most important transformation to come first. Unfortunately, the ....

D. Whitfield, M.L. Soffa, "An Approach to Ordering Optimizing Transformations," 2 nd ACM Symposium on Principles and Practice of Parallel Programming, pp. 137-147, Mar. 1990.

.... The techniques include peephole optimization [McK65] pre defined script based optimization [Bra84] theory based ordering [Wol91] generic probabilistic techniques [Pot94, Cha95] bottleneck identification and elimination approaches, and approaches using enabling and disabling transformations, [Whi90, Pot92, Hua93]. The key novelties of this work are 1) a combination of quantitative and qualitative design advice is given, 2) design advice that is specific to the design at hand is given, 3) the designer remains an integral part of the design process, 4) the approach is extendible to include new ....

Whitfield, D. and Soffa, M.L. An approach to ordering optimizing transformations. ACM Symposium on Principles and Practice of Parallel Programming, 137-147, 1990.

....Carter and Professor Jeanne Ferrante, Co Chairs Successful program optimization requires analysis of profitability. From this analysis, a compiler or runtime system can decide where and how to apply an assortment of program transformations. This two faced problem is called transformation guidance [110, 100, 11, 63, 108]. We consider the desired goal of robust guidance of performance optimizations for hierarchical systems. A guidance system is robust if it unifies disparate sources of knowledge, and makes reasonable decisions hold up, despite a lack of definitive information. In particular, we seek to address ....

....for annotations ranging from dataflow analysis (e.g. which global variables each routine uses) to performance annotations (e.g. how a routine traverses its arrays) Interactions and Staging: We discuss implications of our work on optimization staging in future work, Chapter 10. Whitfield and So#a [110], Pollock and So#a [86] Berson, Gupta and So#a [11, 12] and Cho et al. 23] have also studied this problem: how do the stages of a compiler (such as high level optimizations versus low level optimizations, or register allocation versus instruction scheduling) interact, and how can a compiler ....

[Article contains additional citation context not shown here]

Debbie Whitfield and Mary Lou So#a. An approach to ordering optimizing transformations. In Principles and Practice of Parallel Programming, pages 137--146, Seattle, WA, March 1990.

....specifications can be given modulo associativity and commutativity. The modular forms in gpas not only allow for a more expressive means of specifying transformations modulo the commuting properties of commuting operators, but also alleviates some of the so called phase ordering problems [155]. The phase ordering problem is the problem of ordering the application of transformations in a non confluent TRS to obtain the required transformed result. In this respect, the restructuring and parallelizing compiler transformations are a notorious example of this problem, because it is in ....

....as a commuting property of the commuting loop operator. At first, it may seem odd to have a loop construct that is (self )commuting. However, the advantage of algebraic restructuring of programs is not only the formal treatment of programs, but also the alleviation of the phase ordering problem [155] of the application of restructuring compiler transformations. In fact, the loop transformations implemented in Ctadel s restruct rulebase make extensively use of self commuting loop operators. In this way, aggressive loop fusion is possible, because nested loops are implicitly interchanged in ....

D. Whitfield and M.L. So#a, An Approach to Ordering Optimizing Transformations, in proceedings of the 2 nd ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, 1990, pp. 137--146.

....that our method is much more intuitive for the average programmer because it relies on the familiar concept of graphs. Only few approaches are known which integrate analyses and transformations. SPECIFY [Koc92] additionally provides a proof language but was never implemented completely. GENESIS [WS90] provided one of the starting points of this work. It allows powerful transformation specifications. Preconditions can be specified in a variant of first order logic. However, because fixpoint computations cannot be specified, generation of data flow analysis is not possible. Also the ....

D. Whitfield and M. L. Soffa. An approach to ordering optimizing transformations. In ACM Conference on Principles and Practice of Parallel Programming (PPOPP), 1990.

....to generate specific components: lexical analyzers, parsers, optimizers, parallelizers, and code generators. Unfortunately, there are key components that existing tools cannot directly produce. Further, only recently have researchers focused on the problems of integrating some components [WS90] WS91] Far92] This dissertation proposes a framework for the automatic construction of complete compilers, with special attention to the problems associated with construction of compilers targeting machine that employ various types of hardware parallelism. Integrating many components is a ....

D. Whitfield and M. L. Soffa. An Approach to Ordering Optimizing Transformations. In Proceedings of the Second ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, pages 137--146, Seattle, Washington, March 1990.

....of a program and hence, the efficacy of a transformation may depend on the target architecture. Moreover, some transformations may interfere with others, i.e. transforming a program in one way may make it impossible to apply a transformation that was originally applicable (see, for example, [167]) It will therefore be necessary to identify the order in which Technical Annex PE7195 ACCLAIM 29 transformations are to be applied. We will also investigate techniques for analysing the cost of a program, possibly taking (some aspects of) the target architecture into account. We have made some ....

D. Whitfield and M.L. Soffa. An approach to ordering optimizing transformations. In Proc. Second SIGPLAN Symp. on Principles and Practice of Parallel Programming, 1990.

....have looked to unimodular matrix transformations as a more elegant approach [DOWL90] BANJ90] but there exist legal loop transformations which are not unimodular and thus this approach is limited. A major difficulty in applying transformations is determining in which order to apply them. In [WHIT90] and CHAPTER 1. INTRODUCTION 13 [WOLF90c] this issue is discussed, but in general the relative merit of each transformation depends on the target architecture involved and in which phase of the compiler the transformations can take place. Because of the problems of ordering in ad hoc ....

Whitfield D. and Soffa M.L., "An Approach to Ordering Optimizing Transformations", Proceedings of the 2nd ACM SIGPLAN Symposium on Principles and Practise of Parallel Programming, pp177-186, Seattle 1990.

....of induction variables (i.e. integer variables indexing arrays that are not a loop variable of a surrounding for loop) by a term depending only on loop variables, and ffl eliminating dead code. These transformations are applied in this order just once (regarding ordering of transformations, see [WS90] 3.3 Patterns, Templates, and the Pattern Hierarchy Graph Each nontrivial pattern m is a pair (f m ; m ) consisting of a specification fm of a (mathematical) operation, and a list m of specifications of the types and the data structures of the parameters occurring in fm . For instance, the ....

Debbie Whitfield and Mary Lou Soffa. An Approach to Ordering Optimizing Transformations. In ACM SIGPLAN Programming Language Design and Implementation, pages 137--146, 1990.

....Inherent to the application of program transformations in an optimizing or restructuring compiler is the so called phase ordering problem , i.e. the problem of finding an effective order in which particular transformations must be applied. This problem is still an important research topic [WS90]. An important step forwards in solving the phase ordering problem has been accomplished by the observation that any combination of the iteration level loop transformations loop interchanging, loop skewing and loop reversal (see e.g. AK87, Ban93, PW86, Pol88, Wol86, Wol88, Wol89, Zim90] can be ....

Debbie Whitfield and Mary Lou Soffa. An approach to ordering optimizing transformations. In Proceedings of the second ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, pages 137--146, 1990.

....computation is concentrated in nested loops, many parallel optimizations focus on loops, such as loop distribution, loop fusion, loop interchanging, strip mining, supernode partitioning, and so on. A recent paper by Whitfield and Soffa described an approach for ordering optimizations in a compiler [WhS90]; they attempt to treat scalar and parallel optimizations in a uniform manner, and show that some optimizations work at cross purposes. In particular, some optimizations prevent (disable) application of other optimizations. From this observation they offer a framework from which to deduce an ....

.... 1) endfor S 2# : if( X 0 ) then for I = 1 to N do S 3# : B(I 1) X (B(I) 1) endfor endif Moreover, if we define IF Sinking as IFU 1# , then sinking IFs into loops can enable loop fusion. IFU is related to general ICM, although it is applied to a control construct. Whitfield and Soffa [WhS90] claim ICM can disable FUS, by making the loops nonadjacent; if we assume the compiler works from a dependence structure rather than a lexical structure, ICM will not disable FUS unless there is a dependence chain from the first loop through the invariant code to the second loop. Usually this will ....

[Article contains additional citation context not shown here]

D. Whitfield and M. L. Soffa, An Approach to Ordering Optimizing Transformations, in Proceedings of the Second ACM SIGPLAN Symposium on Principles & Practices of Parallel Programming, March 1990, 137-146.

....the transformation is legal, and the effect of applying a sequence of transformations is determined by the product of the corresponding matrices. Algorithms for handling different kinds of loop nests and loop transformations have been developed by others [24, 36, 37, 42] Whitfield and Soffa [46] developed a framework that unifies the specification of classical and parallelizing optimizations with the goal of examining the interactions between transformations and aiding in the ordering of optimization phases. The framework is based on an axiomatic specification of optimizations using ....

....costs to assignment statements to guide elimination of partial redundancies [26] and characteristics of loop structures [25] such as nesting level and dependences. The more unified approaches take a broader look, considering the use of an axiomatic scheme to represent and analyze characteristics [46] and a matrix based theory of representation and analysis [41, 47] Most of the program characteristics that are used are gathered through static program analysis typically performed by an optimizing compiler. 2.3 Performance Characteristics An execution profiler is a very common tool that is ....

Debbie Whitfield and Mary Lou Soffa. An approach to ordering optimizing transformations. In ACM SIGPLAN Symposium on Principles and Practices of Parallel Programming, pages 137--146, March 1990.

No context found.

Deborah Whitfield and Mary Lou Soffa, `An approach to ordering optimizing transformations', Proc. Second ACM SIGPLAN Symposium on Principles & Practices of Parallel Programming, March 1990, pp. 137--146.

....difficulties for the user when interacting with the parallelizing software. The user needs system support and tools to facilitate the understanding and effective use of code transformations. Although numerous parallelizing tools have been designed and implemented, few provide any form of guidance [2, 7, 13]. Applying a transformation does not always guarantee a time or space benefit. Also, the application of a ############### Partially supported by NSF under Grants CCR 9109089 and IRI 9002180 to the University of Pittsburgh transformation may invalidate conditions for another transformation ....

....does not always guarantee a time or space benefit. Also, the application of a ############### Partially supported by NSF under Grants CCR 9109089 and IRI 9002180 to the University of Pittsburgh transformation may invalidate conditions for another transformation that may be more beneficial [13]. Due to the interactions between schedulers and transformations, the application of a transformation may produce a worse schedule than before the transformation was applied [12] Once the performance evaluation of transformations is determined, the user may need to undo, or remove, ....

D. Whitfield and M. L. Soffa, "An Approach to Ordering Optimizing Transformations," in Proceedings of the Second ACM SIGPLAN Symposium on Principles & Practices of Parallel Programming, pp. 137-146, March 1990.

....compilers and are used to restructure code to enable the exploitation of parallelism in programs. Applying a transformation does not always guarantee a time or space benefit. Also, the application of a transformation may invalidate conditions for another transformation which may be more beneficial [20]. Due to the interaction between schedulers and transformations, the application of a transformation may produce a worst schedule than before the transformation was applied [19] Because it is not clear whether or not a transformation will be effective when it is applied, it may be necessary to ....

....conclusions are drawn in Section 6. 2. Background Interactions of code transformations may occur by one transformation enabling the application of another transformation that previously could not be applied, or one transformation disabling conditions that exist for another transformation [13, 20]. If conditions of a transformation t j are enabled or disabled by another transformation t i , t i is defined to be an affecting transformation to t j and t j is defined as an affected transformation by t i . Enabling interactions among transformations occur when performing a transformation ....

[Article contains additional citation context not shown here]

D. Whitfield and M. L. Soffa, "An Approach to Ordering Optimizing Transformations," in Proceedings of the Second ACM SIGPLAN Symposium on Principles & Practices of Parallel Programming, pp. 137-146, March 1990.

....to use, how to order them, and what parts of a program should be transformed. Applying a transformation does not always guarantee that we can get any time or space benefit. Also, the application of a transformation may invalidate conditions for another transformation which may be more beneficial [26]. Due to the interaction between schedulers and transformations, the application of a transformation may produce a worst scheduling than before transformations are made [25] The user may want to remove these ineffective or inappropriate transformations while maintaining beneficial ....

....[1,6] 1,4] 5,8] 1,8] D) R 3 R 2 Figure 3. A VPDG representation. 4. Specification of Optimizations Traditionally, the approach taken in the development of an optimizer consists of creating algorithms to perform particular optimizations. An alternate approach advocated by Whitfield and Soffa [26, 27] is to specify the components of an optimization using GOSpeL, and then to automatically generate an optimizer from the specifications. The specification of optimizations includes the specification of conditions that exist prior to application, actions to perform the optimization, and conditions ....

D. Whitfield and M. L. Soffa, "An Approach to Ordering Optimizing Transformations," in Proceedings of the Second ACM SIGPLAN Symposium on Principles & Practices of Parallel Programming, pp. 137-146, Mar. 1990.

No context found.

D. Whitfield and M. Soffa, "An approach to ordering optimizing transformations, " ACM Symposium on Principles and Practice of Parallel Programming, Vol. 25, No. 3, pp. 137-146, 1990.

No context found.

D. Whitfield & M. L. Soffa, "An Approach to Ordering Optimizing Transformations," Proceedings 2nd ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, SIGPLAN Notices 25 (Mar. 1990).

No context found.

D. Whitfield and M. L. So#a. An approach to ordering optimizing transformations. In PPOPP'90, pages 137--146, 1990.

No context found.

D.Whitfieldand M.L.Soffa, "An Approach to Ordering Optimizing Transformations", 2nd ACM Symposium on Principles and Practice of Parallel Programming, pp.137-147, March 1990.

First 50 documents

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