## Irregular Coarse-Grain Data Parallelism Under LPARX (0)

Venue: | Journal of Scientific Programming |

Citations: | 17 - 7 self |

### BibTeX

@ARTICLE{Kohn_irregularcoarse-grain,

author = {Scott Kohn and Scott B. Baden},

title = {Irregular Coarse-Grain Data Parallelism Under LPARX},

journal = {Journal of Scientific Programming},

year = {},

volume = {5},

pages = {185--201}

}

### OpenURL

### Abstract

LPARX is a software development tool for implementing dynamic, irregular scientific applications, such as multilevel multilevel finite difference methods and particle methods, on high performance MIMD parallel architectures. It supports coarse grain data parallelism and gives the application complete control over specifying arbitrary block decompositions. LPARX provides structural abstraction, representing data decompositions as first-class objects that can be manipulated and modified at run-time. LPARX, implemented as a C++ class library, is currently running on diverse MIMD platforms, including the Intel Paragon, Cray C-90, IBM SP2, and networks of workstations running under PVM. Software may be developed and debugged on a single processor workstation. 1 Introduction An outstanding problem in scientific computation is how to manage the complexity of converting mathematical descriptions of dynamic, irregular numerical algorithms into high performance applications software. Non-unifo...

### Citations

946 |
Performance FORTRAN Forum. High performance fortran language specification, version 2.0
- High
- 1997
(Show Context)
Citation Context ...programming model that provides run-time support for dynamic, block-irregular data decompositions. General irregular block decompositions are not currently supported by compiled languages such as HPF =-=[21]-=-, Fortran D [19], Vienna Fortran [12], and Fortran 90D [31]. They arise in two important classes of applications: ffl adaptive multilevel finite difference methods [8, 9, 27] that represent refinement... |

713 | PVM: a framework for parallel distributed computing
- Sunderam
- 1990
(Show Context)
Citation Context ...sic message passing support and may run on any MIMD machine. LPARX is currently running on the Intel Paragon, IBM SP2, Cray C-90, single processor workstations, and networks of workstations under PVM =-=[30]-=-. This paper is organized as follows. Section 2 provides an overview of the LPARX programming abstractions. Section 3 describes in detail the parallelization of a simple application, Jacobi relaxation... |

372 |
Adaptive mesh refinement for hyperbolic partial differential equations
- Berger, Oliger
- 1983
(Show Context)
Citation Context ... compiled languages such as HPF [21], Fortran D [19], Vienna Fortran [12], and Fortran 90D [31]. They arise in two important classes of applications: ffl adaptive multilevel finite difference methods =-=[8, 9, 27]-=- that represent refinement regions using block-irregular data structures, and This work was supported by NSF contract ASC-9110793 and ONR contract N00014-93-1-0152. Intel Paragon and Cray C-90 time we... |

351 |
Local Adaptive Mesh Refinement for Shock Hydrodynamics
- Berger, Colella
- 1989
(Show Context)
Citation Context ... compiled languages such as HPF [21], Fortran D [19], Vienna Fortran [12], and Fortran 90D [31]. They arise in two important classes of applications: ffl adaptive multilevel finite difference methods =-=[8, 9, 27]-=- that represent refinement regions using block-irregular data structures, and This work was supported by NSF contract ASC-9110793 and ONR contract N00014-93-1-0152. Intel Paragon and Cray C-90 time we... |

285 | Fortran D Language Specification
- Fox, Hiranandani, et al.
- 1990
(Show Context)
Citation Context ...l that provides run-time support for dynamic, block-irregular data decompositions. General irregular block decompositions are not currently supported by compiled languages such as HPF [21], Fortran D =-=[19]-=-, Vienna Fortran [12], and Fortran 90D [31]. They arise in two important classes of applications: ffl adaptive multilevel finite difference methods [8, 9, 27] that represent refinement regions using b... |

232 |
A partitioning strategy for nonuniform problems on multiprocessors
- Berger, Bokhari
(Show Context)
Citation Context ...ll Theory Center. Portions of this paper are taken from Kohn's Ph.D. dissertation [24]. To appear in J. Scientific Programming 2 ffl parallel computations that require an irregular data decomposition =-=[7]-=- to balance non-uniform workloads across parallel processors, such as particle methods [23]. LPARX hides many of the low-level details, such as interprocessor communication, involved in managing compl... |

138 | Communication optimizations for irregular scientific computations on distributed memory architectures
- Das, Uysal, et al.
- 1994
(Show Context)
Citation Context ...on abstraction which includes a coarse-grain data parallel loop over objects within the collection; pC++ employs a data decomposition scheme similar to that of HPF. The Multiblock PARTI [1] and CHAOS =-=[16]-=- libraries provide run-time support for data parallel compilers such as HPF. CHAOS is targeted towards unstructured calculations such as sweeps over finite element meshes or sparse matrix calculations... |

102 | Distributed pC++: Basic ideas for an object parallel language
- Bodin, Beckman, et al.
- 1993
(Show Context)
Citation Context ...w Vienna Fortran will support dynamic irregular blocking structures such as those required by adaptive mesh refinement and recursive coordinate bisection decompositions. The pC++ programming language =-=[10]-=- implements a collection abstraction which includes a coarse-grain data parallel loop over objects within the collection; pC++ employs a data decomposition scheme similar to that of HPF. The Multibloc... |

86 | Adaptive global optimization with local search
- Hart
- 1994
(Show Context)
Citation Context ...may not be required to rewrite highly optimized Fortran kernels when parallelizing an application. Some current LPARX applications include a 2d geographically structured genetic algorithm application =-=[20]-=-, a 3d smoothed particle hydrodynamics method [24], an adaptive eigenvalue solver for the first principles simulation of real materials [11, 24], and a dimension-independent code for 2d, 3d, and 4d co... |

55 | An Integrated Runtime and Compile-Time Approach for Parallelizing Structured and Block Structured Applications
- Agrawal, Sussman, et al.
- 1995
(Show Context)
Citation Context ...nts a collection abstraction which includes a coarse-grain data parallel loop over objects within the collection; pC++ employs a data decomposition scheme similar to that of HPF. The Multiblock PARTI =-=[1]-=- and CHAOS [16] libraries provide run-time support for data parallel compilers such as HPF. CHAOS is targeted towards unstructured calculations such as sweeps over finite element meshes or sparse matr... |

48 | Extending HPF for Advanced Data Parallel Applications
- Chapman, Mehrotra, et al.
- 1994
(Show Context)
Citation Context ...current vector machines because the programmer must manage computational resources at a very low level. Parallel compiler technology does not yet afford the convenience expected by the user community =-=[13]-=-. While this situation will improve with time, adequate run-time support is essential in applications with dynamic, data-dependent computational structures. We have developed the LPARX parallel progra... |

42 | A Robust Parallel Programming Model for Dynamic, Non-Uniform Scientific Computation - Kohn, Baden - 1994 |

39 |
A method of local corrections for computing the velocity field due to a distribution of vortex blobs
- Anderson
- 1986
(Show Context)
Citation Context ...-consuming portion of the calculation. A naive force evaluation scheme would, for a system of N particles, calculate all O(N 2 ) particleparticle interactions directly. Rapid approximation algorithms =-=[2, 3]-=- accelerate the force evaluation by calculating direct interactions only for those particles lying within a specified cut-off distance. The remaining non-local interactions, which we will not describe... |

32 | ZPL: An Array Sublanguage
- Lin, Snyder
- 1993
(Show Context)
Citation Context ...ons 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 d... |

28 |
FIDIL: A language for scientific programming
- Hilfinger, Colella
- 1989
(Show Context)
Citation Context ...ction specifiers, the Region is a first-class object and may be assigned and manipulated at run-time. The concept of first-class array section objects was introduced in the FIDIL programming language =-=[22]-=-. The Grid is a dynamic array defined over an arbitrary rectangular index set specified by a Region. The Grid is similar to an HPF allocatable array. Each Grid remembers its associated Region, which c... |

28 |
Smoothed particle hydrodynamics, Annual Review of Astronomy and Astrophysics 30
- Monaghan
- 1992
(Show Context)
Citation Context ...approximately one percent to the total execution time of the program. 5.2 Smoothed Particle Hydrodynamics We present performance figures for a 3d smoothed particle hydrodynamics (SPH3D) application 5 =-=[29]-=-. The computational structure of SPH3D is similar to molecular dynamics and other particle calculations which exhibit short-range interactions. Each particle interaction is expensive, requiring approx... |

21 | A Parallel Software Infrastructure for Dynamic Block-Irregular Scientific Calculations
- Kohn
- 1995
(Show Context)
Citation Context ...situation will improve with time, adequate run-time support is essential in applications with dynamic, data-dependent computational structures. We have developed the LPARX parallel programming system =-=[24, 25]-=- to simplify the development of dynamic, non-uniform scientific computations on high performance parallel architectures. Such software support is essential to developing high performance, portable, pa... |

15 |
Object Oriented Implementation of Adaptive Mesh Refinement Algorithms
- Crutchfield, Welcome
- 1993
(Show Context)
Citation Context ... generality. We believe that the programmer should pay the price of such generality only when necessary and advocate the inclusion of more general FIDIL regions as a separate type. Crutchfield et al. =-=[15]-=- independently developed similar abstractions based upon FIDIL for single processor architectures. Based on this framework, they have developed domain-specific libraries for adaptive mesh refinement a... |

13 |
Scalable parallel numerical methods and software tools for material design
- Bylaska, Kohn, et al.
- 1995
(Show Context)
Citation Context ... 2d geographically structured genetic algorithm application [20], a 3d smoothed particle hydrodynamics method [24], an adaptive eigenvalue solver for the first principles simulation of real materials =-=[11, 24]-=-, and a dimension-independent code for 2d, 3d, and 4d connected component labeling for spin models in statistical mechanics [18]. In the following sections, we provide a brief analysis of LPARX overhe... |

10 |
Parallelization of irregularly coupled regular meshes
- Chase, Crowley, et al.
- 1992
(Show Context)
Citation Context ... matrix calculations. Multiblock To appear in J. Scientific Programming 20 PARTI has been targeted to applications with a small number of large, static blocks (e.g. irregularly coupled regular meshes =-=[14]-=-). 7 Conclusions LPARX is a portable programming model and run-time system which supports coarse-grain data parallelism efficiently over a wide range of MIMD parallel platforms. Its structural abstrac... |

10 |
Fortran 90D compiler for distributed memory MIMD parallel computers
- Wu, Fox
- 1991
(Show Context)
Citation Context ...c, block-irregular data decompositions. General irregular block decompositions are not currently supported by compiled languages such as HPF [21], Fortran D [19], Vienna Fortran [12], and Fortran 90D =-=[31]-=-. They arise in two important classes of applications: ffl adaptive multilevel finite difference methods [8, 9, 27] that represent refinement regions using block-irregular data structures, and This wo... |

9 |
Dynamic data distribution
- Chapman, Mehrotra, et al.
- 1993
(Show Context)
Citation Context ...ime support for dynamic, block-irregular data decompositions. General irregular block decompositions are not currently supported by compiled languages such as HPF [21], Fortran D [19], Vienna Fortran =-=[12]-=-, and Fortran 90D [31]. They arise in two important classes of applications: ffl adaptive multilevel finite difference methods [8, 9, 27] that represent refinement regions using block-irregular data s... |

8 |
Flexible communication schedules for block structured applications. (in preparation
- Fink, Baden, et al.
- 1995
(Show Context)
Citation Context ...bal barrier synchronization, which account for the additional message sends. However, it is possible to eliminate this costly synchronization through some simple run-time schedule analysis techniques =-=[17]. The resu-=-lts of these optimizations are reported in the "No Barrier" column of Table 2. LPARX overheads now drop to approximately one percent to the total execution time of the program. 5.2 Smoothed ... |

7 |
A fast vortex method in three dimensions
- Almgren, Buttke, et al.
- 1994
(Show Context)
Citation Context ...-consuming portion of the calculation. A naive force evaluation scheme would, for a system of N particles, calculate all O(N 2 ) particleparticle interactions directly. Rapid approximation algorithms =-=[2, 3]-=- accelerate the force evaluation by calculating direct interactions only for those particles lying within a specified cut-off distance. The remaining non-local interactions, which we will not describe... |

6 |
Portable parallel programming of numerical problems under the LPAR system
- Baden, Kohn
- 1993
(Show Context)
Citation Context ...omposition and interprocessor communication. Whereas FIDIL supports the notion of arbitrary non-rectangular index sets, LPARX restricts index sets to be rectangular. A prototype of LPARX, called LPAR =-=[5]-=-, supported FIDILTo appear in J. Scientific Programming 19 12k 24k 48k 96k Number of Particles 1 10 100 SPH3D Performance Comparison Alpha Cluster (P = 8) Cray C-90 (P = 1) Paragon (P = 16) SP2 (P = 4... |

4 | Parallel cluster identification for multidimensional lattices. (submitted to
- Fink, Huston, et al.
- 1995
(Show Context)
Citation Context ...nvalue solver for the first principles simulation of real materials [11, 24], and a dimension-independent code for 2d, 3d, and 4d connected component labeling for spin models in statistical mechanics =-=[18]-=-. In the following sections, we provide a brief analysis of LPARX overheads and computational results for the smoothed particle hydrodynamics application. Software versions and compiler options for al... |

3 |
Structural abstraction: A unifying parallel programming model for data motion and partitioning in irregular scientific computations. (in preparation
- Baden, Fink, et al.
- 1995
(Show Context)
Citation Context ...d to representing irregular, block-structured decompositions, the concept of structural abstraction is general and extends to other classes of applications, such as unstructured finite element meshes =-=[4]-=-. 2.2 LPARX Data Types LPARX provides the following three abstract data types: ffl Region: an object representing a subset of array index space, ffl Grid: a dynamic array instantiated over a Region, a... |

2 |
The LPARX user's guide v2.0
- Baden, Kohn, et al.
- 1994
(Show Context)
Citation Context ...r. The reduction function mpReduceMax() takes the maximum over all the local values and returns the global maximum value. Various forms of reduction functions are provided by LPARX's standard library =-=[6]-=-. 3.4 FillPatch() The FillPatch() code shown in Figure 7 updates the ghost cell regions of each subgrid with data from the interior (non-ghost cell) sections of adjacent subgrids. For every pair of Gr... |