C. Lin and L. Snyder, "ZPL: An Array Sublanguage," Proc. Worksh on Languages and Compilers for Parallel Computing'93, p96-114, Aug. 1993  Home/Search   Document Details and Download   Summary   Related Articles   Check

....elements of A. Any elements of A that are outside the region are unchanged by the statement. Vector constants are denoted in ZPL using directions, which let a programmer specify relative positions that can perform transformations on regions. A typical declaration might be direction north = [ 1,0]; east = 0,1] There are a number of special operators for transforming or referencing regions. The of operator takes a direction and region as its operands and defines a new boundary region that is adjacent to the base region. The operator, when applied to an array, translates the array s ....

....of A. Any elements of A that are outside the region are unchanged by the statement. Vector constants are denoted in ZPL using directions, which let a programmer specify relative positions that can perform transformations on regions. A typical declaration might be direction north = 1,0] east = [0,1]; There are a number of special operators for transforming or referencing regions. The of operator takes a direction and region as its operands and defines a new boundary region that is adjacent to the base region. The operator, when applied to an array, translates the array s region by a ....

C. Lin and L. Snyder, "ZPL: An Array Sublanguage, " Languages and Compilers for Parallel Computing, 6th Int'l Workshop Proc., Springer-Verlag, Berlin, 1994, pp. 96--114.

....[55] Based on this framework, they have developed domain specific libraries for adaptive mesh refinement applications in gas dynamics [19] Their adaptive mesh refinement libraries have been parallelized using our software infrastructure [141] see Section 3.4. 1) The array sublanguage ZPL [100, 102] employs a form of region abstraction. ZPL does not explicitly manage data distribution, which it assumes is handled by another language. It uses its region constructs to simplify array indexing and as iteration masks; in contrast, LPARX employs Regions to specify run time data decompositions and ....

C. Lin and L. Snyder, ZPL: An array sublanguage, in Proceedings of the Sixth International Workshop on Languages and Compilers for Parallel Computation, Springer-Verlag, 1994, pp. 96--114.

....non uniform parallel memory hierarchies [2] the programmer must judiciously exploit parallelism and locality in the application to match the hardware capabilities. To ease the programmer s burden, programming languages and libraries can hide many low level details of a parallel implementation [20, 24, 16, 1, 12, 10, 35, 15, 11, 23, 28, 4]. We present Kernel Lattice Parallelism (KeLP) a C class library that provides high level abstractions to manage data layout and data motion for dynamic block structured applications. Block structures arise in many scientific applications ranging from finite difference methods for partial ....

Lin, C. and Snyder, L. ZPL:an array sublanguage. In Banerjee, U., Gelernter, D., Nicolau, A., and Padua, D. (Eds.). Languages and Compilers for Parallel Computing, 6th International Workshop Proceedings, Springer-Verlag, New York, 1994, pp. 96--114. to appear in J. Parallel and Distributed Computing

....These are calculations where equations manipulating large arrays of data are transformed into distributed computations by distribution of these arrays across processors. Several modern languages that support data parallel constructs are Fortran 90[7] High Performance Fortran (HPF) 8] and ZPL[9]. ZPL is based on a slightly more general model of computation that identifies Phase Abstractions as a model for parallel computations[10] These Phase Abstractions are identified by the XYZ levels of programming, where the X level corresponds to an individual sequential processor, the Y level ....

....Using domains, it is possible to represent fairly complex grid structures economically, while identifying the array structure allowed for the generation of efficient vector code. The features of domain and map types are recurring themes and appear in other specialized languages such as ZPL[9]. 2.3.2 Strand Strand[28] is a single assignment logic programming based language that is designed for the implementation of parallel applications. A strand program consists of a set of procedures defined by rules. A rule has the form of H : GammaG 1 ; G 2 ; Delta Delta Delta ; Gm jB 1 ; B 2 ....

C. Lin and L. Snyder, "ZPL: An array sublanguage," in Languages and Compilers for Parallel Computing (U. Banerjee and D. Gelernter, eds.), pp. 96--114, Springer-Verlag, 1993.

....framework, they have developed domain specific libraries for adaptive mesh refinement applications in gas dynamics. The LPARX software is currently being employed to support these libraries on parallel computer architectures. A form of region abstraction is used in the programming language ZPL [28], although ZPL s regions are not first class objects; they are used as an execution mask for data parallel computation whereas LPARX abstractions specify data decomposition and express communication dependencies. In contrast to LPARX, High Performance Fortran [21] represents data decompositions ....

C. Lin and L. Snyder, ZPL: An array sublanguage, in Proceedings of the Sixth International Workshop on Languages and Compilers for Parallel Computation, Springer-Verlag, 1994, pp. 96--114.

....programming languages rely on program generation. Prominent examples include software generators (compilers for domain specific languages e.g. see [4] 6] 13] parallelizing compilers (where the emphasis is on transformations of high level abstractions to parallel programs e.g. see [12]) and translators from one high level language to another. Macro expansion itself can be viewed as an example of program generation where the target program coincides with the generator. We believe that program generation, and software generators in particular, will play a key role in future ....

C. Lin and L. Snyder, "ZPL: An Array Sublanguage". In Languages and Compilers for Parallel Computing, U. Banerjee, D. Gelernter, A. Nicolau, and D. Padua, eds, 1993, 96-114.

....non uniform parallel memory hierarchies [2] the programmer must judiciously exploit parallelism and locality in the application to match the hardware capabilities. To ease the programmer s burden, programming languages and libraries can hide many low level details of a parallel implementation [20, 24, 16, 1, 12, 10, 35, 15, 11, 23, 28, 4]. We present Kernel Lattice Parallelism (KeLP) a C class library that provides high level abstractions to manage data layout and data motion for dynamic block structured applications. Block structures arise in many scientific applications ranging from finite difference methods for partial ....

Lin, C. and Snyder, L. ZPL:an array sublanguage. In Banerjee, U., Gelernter, D., Nicolau, A., and Padua, D. (Eds.). Languages and Compilers for Parallel Computing, 6th International Workshop Proceedings, Springer-Verlag, New York, 1994, pp. 96--114.

....substantial impact on a program s performance. Under the message passing programming model, the programmer must manage all aspects of data distribution by hand. Programming languages and run time libraries that manage the lowlevel details of data distribution can alleviate the programmer s burden [10, 9, 8, 12, 1, 4, 11]. Perhaps the simplest strategy for managing distributed data is the mapping array, a many to one mapping of data elements to processors[9] Mapping arrays provide the most flexibility of any data distribution scheme, but incur a large space overhead since one mapping element must be stored for ....

....strategy for aligning distributed arrays relative to each other. 3 Run time primitives for data distribution Geometric primitives were introduced as first class objects in FIDIL [13] as a means for managing index sets of arrays, and are employed by the ZPL language as compiletime iteration masks[12]. The LPARX [11] class library implements geometric domain abstractions as first class run time objects. We shall use the following subset of LPARX primitives to implement dynamic data distribution. 3.1 LPARX primitives A Point is an n tuple representing a point in Z n . Addition and ....

C. Lin and L. Snyder, ZPL: An array sublanguage, in Languages and Compilers for Parallel Computation, 6th International Workshop Proceedings, Springer-Verlag, 1994, pp. 96--114.

....cases such control over the program is not needed or the programmer does not want to use the full generality the system provides. Also, lower level details result in the loss of higher level information that the compiler might be able to explore to perform a better optimization. The ZPL language [Lin and Snyder 1994] provides high level abstractions of aggregate data types and data parallel operations on them. The ZPL programmer conveys high level information to the compiler which enables optimizations, for example, of communication and array accesses. The aggregate data types and operations make the programs ....

....Finally, Section 11 presents our discussion of the limitations of our approach, directions for future work, and conclusions. 2 Overview of the ZPL Language This section is intended for the reader who is not familiar with ZPL. However, it is only a very brief overview. The reader is referred to [Lin 1994] for further details. 2.1 Data Structures All data structures fall into two categories, scalar and parallel. The scalar data structures are those found in Pascal like languages (except for pointers) such as integers, reals, arrays, records, etc. These data structures are processed sequentially. ....

[Article contains additional citation context not shown here]

Lin, C., and Snyder, L. ZPL: An Array Sublanguage. In U. Banerjee et al. (eds.), Languages and Compilers for Parallel Computing, Springer-Verlag, 1994.

....and hence reduce some of the flexibility issues discussed above. Because the mapping is performed by the data parallel run time system, the debugger must understand the way that data is distributed across processes. We have chosen to experiment with a prototype data parallel language called ZPL [10], developed at the University of Washington. ZPL uses a complex data structure at run time called an ensemble to describe the actual run time mapping. Accordingly, special knowledge must be included for each language that is to be supported greatly increasing development and maintenance costs ....

Lin, C. and Snyder, L. "ZPL: An Array Sublanguage", Languages and Compilers for Parallel Computing, U. Banerjee, D. Gelernter, A. Nicolau, and D. Padua, eds, pp. 96-114, 1993.

....develop applications more quickly and more easily by allowing programs to be coded in terms of high level abstractions. The abstractions we are most familiar with are domain independent (e.g. classes, inheritance, parameterization, etc. although domain specific abstractions are just as common [Lin93]. Compilers, in turn, formalize and mechanize the generation of efficient code for such abstractions, thus relieving the burden of programmers to code (and debug) low level software. Compilers are, in effect, existence proofs that certain, well understood concepts need never be hand coded again. ....

C. Lin and L. Snyder, "ZPL: An Array Sublanguage", Languages and Compilers for Parallel Computing, Springer-Verlag, 1993.

....a high performance parallel code, the programmer faces myriad tradeoffs concerning both low level implementation details and highlevel algorithmic decisions. To ease the programmer s burden, programming languages and libraries can hide many low level details of a parallel implementation (e.g. [78, 92, 54, 3, 45, 43, 110, 52, 44, 85, 97, 16]) A useful set of programming abstractions allows the programmer to express efficient algorithms at a high level, while the software infrastructure manages tedious low level details. Many software systems provide specific support for block structured scientific calculations. Block structures ....

....body of data parallel languages, including Fortran D [67] Fortran 90D [32] Kali [100] CM Fortran [132] and Vienna Fortran [47] There have also been several data parallel extensions to C (e.g. 119] Fx [74] combines the data parallel HPF ideas with task parallel constructs. The ZPL language [97] provides powerful geometric semantics to express data parallel computation. Like KeLP, ZPL relies on a Region abstraction to describe rectangular subsets of index space. The ZPL programmer uses Regions to specify iteration space masks, which proves useful for some types of boundary conditions. ....

C. Lin and L. Snyder. ZPL:an array sublanguage. In U. Banerjee, D. Gelernter, A. Nicolau, and D. Padua, editors, Proc. Languages and Compilers for Parallel Computing, 6th Int'l Workshop, pages 96--114. Springer-Verlag, 1994.

....a computation. These values often become arrays when the code is converted to a data parallel language. Figure 2 shows code fragments indicating how a temporary variable in the C code might be promoted to an array written in ZPL, a data parallel language developed at the University of Washington [9]. In ZPL, the array is a fundamental data type and can be used in expressions in a similar manner to scalar values. Arrays are defined in terms of regions, which are similar to sub arrays in Fortran 90 [6] C Code ZPL Code 128 129 130 131 132 133 134 for (i=0; i 100; i ) temp = x[i] y[i] ....

Lin, C., Snyder, L., "ZPL: An Array Sublanguage", Languages and Compilers for Parallel Computing, U. Banerjee, D. Gelernter, A. Nicolau, and D. Padua, eds, 96 -- 114 (1993).

....for they amplify any ineffectiveness or inefficiency in the message passing layer by the multiplicity of the processors at the node. 3 A Hierarchical Communication Model 3. 1 Historical Perspective Current parallel programming languages and libraries represent a wide range of design choices [20, 24, 17, 14, 2, 11, 9, 30, 12, 10, 21, 27, 5], and share the same underlying model of a parallel computer, which we refer to as the single tier multicomputer model (Fig. 1a) This model directly reflects the structure of many commercial distributed memory parallel computers, as well as formal parallel computation models, such as CTA [35] ....

C. Lin and L. Snyder, ZPL:an array sublanguage, in Proc. Languages and Compilers for Parallel Computing, 6th Int'l Workshop, U. Banerjee, D. Gelernter, A. Nicolau, and D. Padua, eds., Springer-Verlag, 1994, pp. 96--114.

....new language was intended for programming the third level of the phase abstractions XYZ levels. For this reason the language was called ZPL, short for the Z level programming language. A seminar in Spring 1993 was devoted to language design, and it was first presented the summer of the same year [LS94b] Ruth Anderson, Bradford Chamberlain, Sung Eun Choi, George Forman, Calvin Lin, Keith Partridge, Larry Snyder and W. Derrick Weathersby attended the seminar. The original ZPL was primitive. Only a small class of data parallel applications were well suited 33 for ZPL implementation, but those ....

Calvin Lin and Lawrence Snyder. ZPL: an array sublanguage. In Sixth International Workshop on Languages and Compilers for Parallel Computing, pages 96-- 114, Berlin, Germany, 1994. Springer--Verlag.

.... Thus, the ZPL programming language was built on the CTA abstract parallel machine model [43] and a scientific goal of ZPL, beyond the three language goals mentioned above, was to test the second half of this conjecture [34, 45] Several core concepts of ZPL were developed and announced in 1993 [35]. The first portability and performance measurements for the base language were reported in 1994 [36] and these were followed by additional comparisons in 1995 [33] The language was substantially enhanced over the next two years, and in July 1997, ZPL was packaged with appropriate support ....

....model in a specific way. Other language designs could interpret it differently. Finally, it should be noted that ZPL is the array subset of a larger language, dubbed Advanced ZPL, which will provide support for general data structures and more general forms of parallelism, such as task parallelism [35]. 3 A Slice of ZPL This section introduces some basic concepts of the ZPL language and explains how these are executed on an abstract parallel machine. As evident in the simple ZPL program shown in Figure 2, ZPL has many of the same data types and control structures found in standard imperative ....

[Article contains additional citation context not shown here]

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Uptal Banerjee, David Gelernter, Alexandru Nicolau, and David Padua, editors, Languages and Compilers for Parallel Computing, pages 96--114. SpringerVerlag, 1993.

....1: Different representations of the same array language computation in (a) RL and ZPL, b) APL, and (c) Fortran 90 and Matlab. puters. Furthermore, we compare the region based representation to the conventional subscripted form. Although the region based ZPL language has been described before [10], this is the first discussion of regions as an abstract programming language concept. This paper is organized as follows. Section 2 describes the advantages of region based programming. In Section 3, we give a formal definition for regions and describe RL s support for region declarations and ....

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Uptal Banerjee, David Gelernter, Alexandru Nicolau, and David Padua, editors, Workshop on Languages and Compilers for Parallel Computing, pages 96--114. SpringerVerlag, 1993.

....identification of index locality, thereby improving programmers understanding of their codes performance on parallel computers. Furthermore, we compare the region based representation to the conventional subscripted form. Although the region based ZPL language has been described before [9], this is the first discussion of regions as an abstract programming language concept. Section 2 describes the advantages of region based programming. In Section 3, we give a formal definition for regions and describe RL s support for region declarations and operators. In Section 4, we introduce ....

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Uptal Banerjee, David Gelernter, Alexandru Nicolau, and David Padua, editors, Workshop on Languages and Compilers for Parallel Computing, pages 96--114. Springer-Verlag, 1993.

....evaluate the effect of data dependences on compiler analysis, we consider two versions of each benchmark: one programmed using DO loops, and the second using F90 constructs and or HPF s Forall statement. For the comparison, we also consider the performance of each benchmark written in MPI and ZPL [16], a data parallel language developed at the University of Washington. Since message passing programs yield scalable performance but are not convenient, the MPI results represent a level of performance that the HPF programs should use as a point of reference. The motivation for including the ZPL ....

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Languages and Compilers for Parallel Computing. 1993.

....the biological structures often form very complex patterns that require large computational domains for their representation. This paper makes two primary contributions. First, we describe two new algorithms for solving mathematical biology problems. Second, by using the ZPL programming language [15] to implement these algorithms, we show that ZPL solves many of the problems that inhibit the widespread use of parallel machines. The first algorithm models the structure of fibroblasts, which are cells of the connective tissue that have been extensively studied by biologists. Fibroblast patterns ....

....pattern. Detailed numerical simulations and bifurcation pattern sequences, as well as non dimensionalized (biological) parameter values, are reported elsewhere [23] 4 The ZPL Array Sublanguage ZPL is a data parallel language that allows arrays and subarrays to be manipulated as whole entities [15]. The language provides constructs that lead to concise programs while eliminating tedious and error prone array indexing. The language s conciseness was first illus trated by the SIMPLE computational fluid dynamics benchmark, which is approximately 5000 lines when coded in C plus message ....

[Article contains additional citation context not shown here]

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Uptal Banerjee, David Gelernter, Alexandru Nicolau, and David Padua, editors, Languages and Compilers for Parallel Computing, pages 96--114. Springer-Verlag, 1993.

....for communicating up and down a tree structure. Reductions, which naturally combine both communication and computation, are notable exceptions where the separation of 3 Since the submission of this paper, an array sublanguage known as ZPL has been developed to support data parallel computations [35, 47, 31, 37]. While its syntax differs significantly from Orca C, ZPL remains true to the Phase Abstractions model. It provides a powerful Z level language that hides all of the X and Y level details from the user. communication from computation is not desirable. For such operations it suffices to define a ....

C. Lin and L. Snyder. ZPL: An array sublanguage. In U. Banerjee, D. Gelernter, A. Nicolau, and D. Padua, editors, Languages and Compilers for Parallel Computing, pages 96--114. Springer-Verlag, 1993.

....use enables the source level identification of index locality, thereby improving programmers understanding of their codes performance. Furthermore, we compare the region based representation to the conventional subscripted form. Although the region based ZPL language has been described before [7], this is the first discussion of regions as an abstract programming language concept. From the trivial example of Figure 1, regions may appear to be a minor syntactic variation on the other forms of indexing. However, region based programming provides a powerful abstraction that has advantages ....

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Uptal Banerjee, David Gelernter, Alexandru Nicolau, and David Padua, editors, Workshop on Languages and Compilers for Parallel Computing, pages 96-- 114. Springer-Verlag, 1993.

....evaluate the effect of data dependences on compiler analysis, we consider two versions of each benchmark: one programmed using DO loops, and the second using F90 constructs and or HPF s Forall statement. For the comparison, we also consider the performance of each benchmark written in MPI and ZPL [16], a data parallel language developed at the University of Washington. Since message passing programs yield scalable performance but are not convenient, the MPI results represent a level of performance that the HPF programs should use as a point of reference. The motivation for including the ZPL ....

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Languages and Compilers for Parallel Computing. 1993.

....ZPL s better performance comes from its clean expression of parallelism that allows for better compiler analysis. 1 Introduction ZPL is a new data parallel array language designed at the University of Washington to provide high performance parallel computation for all MIMD parallel computers [7]. HPF was recently defined by the High Performance Fortran Forum to provide data parallel programming and top performance on MIMD and SIMD computers [4, p.1] With such similar ambitions, it is natural to compare the two languages. This paper compares ZPL and HPF, concentrating on performance ....

....of writing new parallel programs. The evidence presented will favor ZPL over HPF, for the programs will likely run faster and be easier to understand and maintain. As a crude measure of readability, the SIMPLE benchmark is 2400 lines of Fortran 77 without HPF directives but only 500 lines in ZPL [7]. The discussion in Section 4 illustrates this point more clearly. In cases where a Fortran program exists, our experience suggests that the effort to understand it well enough to select the proper HPF annotations can be comparable to the effort required to rewrite it in ZPL. Moreover, after ....

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Uptal Banerjee, David Gelernter, Alexandru Nicolau, and David Padua, editors, Languages and Compilers for Parallel Computing, pages 96--114. SpringerVerlag, 1993.

.... we empirically evaluate these optimizations using four benchmark programs for two modern parallel machines, the Intel Paragon and the Cray T3D and two communication mechanisms, message passing and one way communication (i.e. T3D s SHMEM libraries) These benchmark programs are written in ZPL [17, 18], a portable data parallel array language similar to the array subset of Fortran 90. The compiler supports a instrumented compiler where optimizations are performed in a machine independent manner (ZPL source programs are compiled to SPMD ANSI C and linked with machine dependent libraries) ....

Calvin Lin and Lawrence Snyder. ZPL: An array sublanguage. In Uptal Banerjee, David Gelernter, Alexandru Nicolau, and David Padua, editors, Workshop on Languages and Compilers for Parallel Computing, pages 96--114. Springer-Verlag, 1993.

No context found.

C. Lin and L. Snyder, "ZPL: An Array Sublanguage," Proc. Worksh on Languages and Compilers for Parallel Computing'93, p96-114, Aug. 1993

No context found.

C. Lin and L. Snyder, "ZPL: An Array Sublanguage," Languages and Compilers for Parallel Computing , U. Banerjee, D. Gelernter, A. Nicolau, and D. Padua, eds, pp. 96-114, 1993.

No context found.

C. Lin and L. Snyder. ZPL: An Array Sublanguage. Proceedings of the Languages and Compilers for Parallel Computing Conference, pages 96--114, 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