W. G. Vree and P. H. Hartel. Communication lifting: fixed point computation for parallelism. J. functional programming, 5(4):549-- 581, Oct 1995. ftp:// ftp.wins.uva.nl/ pub/ computer-systems/ functional/ reports/ JFPcommunicationlifting.ps.Z.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... Southampton team developed sequential compiler technology [HGW89, GHW90] and performance analysis techniques [HGW91, HaLa92] Researchers at Imperial college and Amsterdam used the Southampton compiler as a basis for investigating various parallel implementation techniques [CHK 92, CHK 93, VrHa92] An important component of the FAST system is a highly optimising compiler for Haskell 1.0 [HuWa90] on a single transputer [GHW90] This compiler, which underlies the parallel implementation (see below) contains a number of analyses and syntheses specified within a single 3.4.3 Other ....

WG Vree and PH Hartel, Communication Lifting: Fixed Point Computation for Parallelism. Technical Report CS-92-07, Department of Computer Systems, University of Amsterdam, December 1992.

.... fournissant, dans un cadre strictement fonctionnel, des constructions apparent ees a celles de langages imp eratifs a la Esterel [4] D autre part, ce travail est clairement reli e aux travaux sur la d eforestation [18, 19] Mais le travail le plus proche est sans doute celui de Vree et Hartel [17]. Cependant, leur approche ne s applique pas a des fonctions ne pr eservant pas la longueur, telles que merge , qui sont pourtant centrales dans notre approche. Enfin, il faut noter le travail de Fr ed eric Boussinot sur les r eseaux r eactifs [5] qui d ecrit de meme des r eseaux synchrones, ....

W.G. Vree and P.H. Hartel. Communication lifting : fixed point computation for parallelism. J. Functional Programming, 1(1):1-- 33, 1993.

....functional system should provide a good starting point for the Prolog annotations. These rather low level annotations have been encapsulated in higher level skeletons. A synchronous process network skeleton has recently been developed for our parallel functional language (see Vree and Hartel [VH94] where the synchronicity has been exploited to optimise the implementation of the skeleton. Research into a classification of asynchronous process networks and the exploitation of properties of classes of such networks is a logical extension of this work. The skeletons that have been found ....

....transformation tool set will thus rely on the fact that the user has identified opportunities for the introduction of specific parallel skeletons. The skeletons will include Prolog versions of the skeletons for which we have already optimised our parallel functional system (see Vree and Hartel [VH94] It will thus be necessary to investigate the relation between the parallel semantics of Prolog and the parallel semantics of the functional system. Once the skeletons have been introduced (in the form of annotations or relations) the tools will enable the user and or the annotations to make ....

W. G. Vree and P. H. Hartel. Communication lifting: fixed point computation for parallelism. J. functional programming (to appear, also Technical report CS-92-07, Dept. of Comp. Sys, Univ. of Amsterdam), 1994.

....extended works on the subject, and it is quite difficult to summarize all of them. Yet, roughly speaking, these can be classified into two groups: ffl Particular methods, based on some set of special purpose primitives, among which we can cite deforestation shortcuts [3] communication lifting [24], and the one presented here. It is quite difficult to compare them, because each has its proper set of primitives and constructs as well as its own goal. The communication lifting method [24] is very similar to that can be achieved with Lustre, and thus doesn t intend to deal with dynamical ....

....of special purpose primitives, among which we can cite deforestation shortcuts [3] communication lifting [24] and the one presented here. It is quite difficult to compare them, because each has its proper set of primitives and constructs as well as its own goal. The communication lifting method [24] is very similar to that can be achieved with Lustre, and thus doesn t intend to deal with dynamical networks. Furthermore, it hardly deals with non length preserving functions like filter. On the contrary, our when primitive is a filter like, non length preserving, function, which is central to ....

[Article contains additional citation context not shown here]

W.G. Vree and P.H. Hartel. Communication lifting : fixed point computation for parallelism. Journal of Functional Programming, 1(1):1--33, 1993.

....books on the subject. The lack of parsing facilities makes inputting abstract syntax inconvenient. An elegant way to add scanning and parsing facilities would be to use parser combinators [14] This could be achieved without adding further facilities to latos, see for example our earlier work [29]. With the abstract syntax and a sample code fragment in place it is now time to look at the semantics proper. 3.2 Semantic functions The semantic function A below defines the value of arithmetic expressions (a) An expression may contain identifiers (x) so, the semantic function must know ....

W. G. Vree and P. H. Hartel. Communication lifting: fixed point computation for parallelism. J. functional programming, 5(4):549-- 581, Oct 1995. ftp:// ftp.wins.uva.nl/ pub/ computer-systems/ functional/ reports/ JFPcommunicationlifting.ps.Z.

....application programs tend to hide the parallelism, and are therefore not trivially annotated. For these programs several transformations have been designed that restructure programs with other types of parallelism into programs suitable for sandwich annotations. ffl Communication lifting [11] transforms synchronous process networks [12] used for geometric parallelism and synchronous pipelines) into a divide and conquer structure. This opens a large class of applications. ffl The grain size transformation [10] can be used to enlarge or reduce the grain size of parallel computations ....

....whole will compute generation after generation, while managing only a single stream of tuples. Within a generation, the original stream elements are now accessible to the functions that used to operate on stream elements as the elements of one large tuple. The communication lifting transformation [11] has been developed to change a synchronous network into such a joint management form. Figure 8 presents the result of the transformation; a semi formal derivation of this result is given in the appendix. The transformation is described in full in our earlier work [11] where we also present a ....

[Article contains additional citation context not shown here]

W. G. Vree and P. H. Hartel. Communication lifting: fixed point computation for parallelism. J. functional programming, 5(4):549--581, Oct 1995.

....In the past we have worked on parallel functional solutions to the discretised shallow water equations [27] The solution method is based on domain decomposition of nonoverlapping boxes; it also computes exact solutions. We have developed a number of ways to effect the exchange of borders [13, 31, 32]. In the present paper we will again use the shallow water equations as a case study. Section 2 describes the partitioning of a grid and the associated problems involved in the exchange of borders. Section 3 describes a declarative solution for the exchange of borders. This solution is specific to ....

W. G. Vree and P. H. Hartel. Communication lifting: fixed point computation for parallelism. J. functional programming (to appear, also Technical report CS-9207, Dept. of Comp. Sys, Univ. of Amsterdam), 1994.

....books on the subject. The lack of parsing facilities makes inputting abstract syntax inconvenient. An elegant way to add scanning and parsing facilities would be to use parser combinators [13] This could be achieved without adding further facilities to latos, see for example our earlier work [29]. With the abstract syntax and a sample code fragment in place it is now time to look at the semantics proper. 3.2 Semantic functions The semantic function A below defines the value of arithmetic expressions (a) An expression may contain identifiers (x) so, the semantic function must know about ....

W. G. Vree and P. H. Hartel. Communication lifting: fixed point computation for parallelism. J. functional programming, 5(4):549--581, Oct 1995.

....of the type signatures. solid 605 [5] list, tuple, data, float Point membership classification algorithm from a solid modeling library for computational geometry. complab 653 [18] list, tuple, data Image processing application that labels connected regions of pixels as objects. transform 834 [23] list, tuple, data Transformation of a number of programs represented as synchronous process networks into master slave style parallel programs. parstof 1192 [4] list, tuple, data Lexing and parsing based on Wadler s parsing method [24] of a medium size program. arrays = uses arrays float = uses ....

W. G. Vree and P. H. Hartel. Communication lifting: fixed point computation for parallelism. J. Functional Programming (to appear, also Technical report CS-9207, Dept. of Comp. Sys, Univ. of Amsterdam), 1994.

....be implemented successfully on such a machine, another set of transformations has been developed to turn algorithms based on process networks into divide and conquer programs. The basis of the transformation from a cyclic process network to a divide and conquer algorithm is communication lifting [68]. Consider as an example the set reset flipflop as shown in Figure 3. The arrows represent streams of bits, which can be implemented as infinite lists of 0s and 1s, as in the Miranda program below: set (cs) reset (ds) Gamma Gamma Gamma Gamma Gamma Gamma Gamma nand 1 ....

....be connected in a cyclic fashion. The communication lifting transformation in effect separates the communication aspect from the computation of the next states. The communication lifting transformation has been formally specified and also implemented as an automatic program transformation tool [68]. The end result of the transformation is: setreset1 cs ds = tl (map 1 sel 3 (maps 2 nextstate s0 cs ds) where s0 = dummy output,0,1) dummy output = 1 nextstate (x,a,b) c d = a,a ,b ) where a = hh bnot (band d b) ii b = hh bnot (band a c) ii sel 3 (a,b,c) a map ....

[Article contains additional citation context not shown here]

W. G. Vree and P. H. Hartel. Communication lifting: fixed point computation for parallelism. J. functional programming, 5(4):549--581, Oct 1995.

No context found.

W.G. Vree and P.H. Hartel. Communication lifting : fixed point computation for parallelism. J. Functional Programming, 1(1):1--33, 1993.

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