Gail Alverson, William Griswold, David Notkin, and Lawrence Snyder. A flexible communication abstraction for nonshared memory parallel computing. In Proceedings of Supercomputing '90, November 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....so they can be run on machines of any size. Programming languages are also necessary for writing fully parallel programs. Functional programming, parallel Fortran D [HIRA92] PRAM languages (as suggested in [VALI90b] and message passing languages such as OCCAM [INMO84] and Ensembles [GRIS90][ALVE90] are all attempts to provide user control of parallelism. There are architecture independent languages (PICL [GEIS90] Data Parallel C [HATC91] proposed parallel programming environments (C Linda [CARR90] C and Presto [BERS88] Amber [CHAS89] new types of approaches for parallelism ....

G. A. Alverson, W. G. Griswold, D. Notkin, and L. Snyder, "A Flexible Communication Abstraction for Nonshared Memory Parallel Computing," in Proceedings of Supercomputing `90, 1990.

....rich and interesting languages, it has not always served the goals of simplicity, cleanliness or efficiency. ZPL is a different kind of language. ZPL is an array sublanguage of the Orca family of parallel programming languages [11, 12] The Orca languages provide a general MIMD programming model [2, 6, 17], allowing programmers to write efficient, portable and scalable parallel programs. For many tasks, such as initializing an array A to all 0 s, a full MIMD programming model is overly general. For other tasks, such as summing the elements of the array A, the best parallel solution is generally ....

.... use Region R2 for this function call end The above uses of regions assume that all array operands have been declared to include at least the points in their respective regions. Arrays can be declared using regions as shown below: var A: real [R] B: real [R] C: real [R] 9 NE2 = [ 2,2] SE2 = 2, 2] east2 = 0, 2] SE2 of R] east2 of R] NE2 of R] Figure 1: Illustration of Of Regions. By applying regions to entire statements, ZPL programs are syntactically clean since they are not muddled by complicated indices. Significantly, region scopes are the only way to ....

[Article contains additional citation context not shown here]

Gail Alverson, William Griswold, David Notkin, and Lawrence Snyder. A flexible communication abstraction for nonshared memory parallel computing. In Proceedings of Supercomputing '90, November 1990.

....parameters, recursion, control structures, etc. These are useful abstractions that are not provided by the machine model directly but are important for practical programming. The phase abstractions programming model is a parallel analog of the imperative procedural programming model [Sny90, AGNS90] providing equivalent features as well as data allocation and processor assignment information. The programming model defines a scalable unit of parallelism that encapsulates three aspects of parallel computations code, data and communication so that performance critical characteristics of a ....

.... memory view was taken, because it more accurately abstracted the great diversity of machines that existed [AS91, NS92] Snyder believed automatic parallelization and a shared memory model were not the solution, so he and his colleagues developed the phase abstractions programming model [Sny90, AGNS90] Phase abstractions provides a framework for structuring parallel programs, but it is not itself a programming language. Snyder and his students began experimenting with programming techniques and language abstractions founded on phase abstractions. David Socha s thesis developed Spot, the first ....

Gail A. Alverson, William G. Griswold, David Notkin, and Lawrence Snyder. A flexible communication abstraction for nonshared memory parallel computing. In Proceedings of Supercomputing '90, pages 584--93, 1990.

....realistic for all existing medium and large scale parallel computers. The Phase Abstractions extend the CTA in the same way that the sequential imperative programming model extends the von Neumann model. The main components of the Phase Abstractions are the XYZ levels of programming and ensembles [1, 19, 43]. 2.1 XYZ Programming Levels A programmer s problem solving abilities can be improved by dividing a problem into small, manageable pieces assuming the pieces are sufficiently independent to be considered separately. Additionally, these pieces can often be reused in other programs, saving time ....

....parallel implementations of the FFT, matrix multiplication, matrix transposition, sort, and global maximum. A phase has a characteristic communication structure induced by the data dependencies among the processes. For example, the FFT induces a butterfly, while Batcher s sort induces a hypercube [1]. Finally, the Z level corresponds to the actions of the CTA s global controller, where sequences of parallel phases are invoked and synchronized. A Z level program gives the high level logic of the computation by specifying the sequential invocation of phases (although their execution may ....

[Article contains additional citation context not shown here]

G. Alverson, W. Griswold, D. Notkin, and L. Snyder. A flexible communication abstraction for nonshared memory parallel computing. In Proceedings of Supercomputing '90, November 1990.

....more recent BSP [48] and LogP [8] models present a similar view of a parallel machine and for the most part suggest a similar way of programming parallel computers. model extends the von Neumann model. The main components of the Phase Abstractions are the XYZ levels of programming and ensembles [1, 19, 46]. 2.1 XYZ Programming Levels A programmer s problem solving abilities can be improved by dividing a problem into small, manageable pieces assuming the pieces are sufficiently independent to be considered separately. Additionally, these pieces can often be reused in other programs, saving time ....

....parallel implementations of the FFT, matrix multiplication, matrix transposition, sort, and global maximum. A phase has a characteristic communication structure induced by the data dependencies among the processes. For example, the FFT induces a butterfly, while Batcher s sort induces a hypercube [1]. Finally, the Z level corresponds to the actions of the CTA s global controller, where sequences of parallel phases are invoked and synchronized. A Z level program specifies the high level logic of the computation and the sequential invocation of phases (although their execution may overlap) that ....

[Article contains additional citation context not shown here]

G. Alverson, W. Griswold, D. Notkin, and L. Snyder. A flexible communication abstraction for nonshared memory parallel computing. In Proceedings of Supercomputing '90, November 1990.

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