Rafael Asenjo, Eladio Gutierrez, Yuan Lin, David Padua, Bill Pottenger, and Emilio Zapata. On the Automatic Parallelization of Sparse and Irregular Fortran Codes. Technical Report 1512, Univ. of Illinois at Urbana-Champaign, Center for Supercomputing Res. & Dev., December 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Because the access pattern of the tested array is sparse, the case can be handled as a regular reduction array or as a shared array. When handled as a regular reduction array, reduction optimization (PCLR) has been applied. For the loop move3 goto100 in Dsmc3d privatization removes all dependences [1] but, due to its sparse nature, causes high initialization and final merging overhead. Spark98 is a sparse matrix and dense vector multiplication C kernel. Rmv needs only reduction optimization. 6.1 Evaluation We have used Spice for reduction verification and Spice, Euler, and Rmv for evaluating ....

R. Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottenger, and E. Zapata. On the automatic parallelization of sparse and irregular fortran codes. Technical Report 1512, Center for Supercomputing Research and Development, February 1997.

....been applied. If it is handled as a shared array, the loop becomes partially parallel and will cause a dependence related failure just before it finishes. After recovery, the remaining iterations are executed sequentially. For the loop move3 goto100 in Dsmc3d privatization removes all dependences [1] but, due to its sparse nature, causes high initialization and final merging overhead. By treating the tested array as a shared array and applying SCA we obtain good overall results. Spark98 is a sparse matrix and dense vector multiplication C kernel. Rmv needs only reduction optimization. 4.3 ....

R. Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottenger, and E. Zapata. On the automatic parallelization of sparse and irregular fortran codes. Tech. Rept. 1512, Center for Supercomputing Research and Development, February 1997.

....of this loop presents an outer sequential loop and inner parallel loop that uses sparse reduction parallelization. On this inner loop we have applied the reduction optimization algorithm and obtained very good results. For the loop move3 goto100 in Dsmc3d, privatization removes all dependences [1] but, due to its sparse nature, causes high initialization and final merging overhead. By treating the tested array as a shared array and applying the UPAR algorithm we obtain good overall results. Spark98 is a sparse matrix and dense vector multiplication C kernel. Rmv needs only reduction ....

R. Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottenger, and E. Zapata. On the automatic parallelization of sparse and irregular fortran codes. Technical Report 1512, Center for Supercomputing Research and Development, February 1997.

....Becausethe access pattern of the tested array is sparse, the case can be handled as a regular reduction array or as a shared array. When handled as a regular reduction array, reduction optimization (PCLR) has been applied. For the loop move3 goto100 in Dsmc3d privatization removes all dependences [1] but, due to its sparse nature, causes high initialization and final merging overhead. Spark98 is a sparse matrix and dense vector multiplication C kernel. Rmv needs only reduction optimization. 5.3 Evaluation We have used Spice for reduction verification and Spice, Euler, and Rmv for evaluating ....

R. Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottenger, and E. Zapata. On the automatic parallelization of sparse and irregular fortran codes. Technical Report 1512, Center for Supercomputing Research and Development, February 1997.

No context found.

Rafael Asenjo, Eladio Gutierrez, Yuan Lin, David Padua, Bill Pottenger, and Emilio Zapata. On the Automatic Parallelization of Sparse and Irregular Fortran Codes. Technical Report 1512, Univ. of Illinois at Urbana-Champaign, Center for Supercomputing Res. & Dev., December 1996.

No context found.

Rafael Asenjo, Eladio Gutierrez, Yuan Lin, David Padua, Bill Pottenger, and Emilio Zapata. On the Automatic Parallelization of Sparse and Irregular Fortran Codes. Technical Report 1512, Univ. of Illinois at Urbana-Champaign, Center for Supercomputing Res. & Dev., December 1996.

No context found.

R. Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottengerg and E. Zapata, On the Automatic Parallelization of Sparse and Irregular Fortran Codes, Technical Report 1512, University for Illinois at Urbana-Champaign, Center for Supercomputing R&D., December 1996.

No context found.

Rafael Asenjo, Eladio Gutierrez, Yuan Lin, David Padua, Bill Pottenger, and Emilio Zapata. On the Automatic Parallelization of Sparse and Irregular Fortran Codes. Technical Report 1512, Univ. of Illinois at Urbana-Champaign, Center for Supercomputing Res. & Dev., December 1996.

No context found.

R. Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottenger, and E. Zapata. On the automatic parallelization of sparse and irregular fortran codes. Technical Report CSRD-TR-1512, Dept. of Computer Science, University of Illinois at Urbana Champaign, December 1996.

....4 discusses some newly identified transformation techniques, the effectiveness of which is evaluated in Sect. 5. Section 6 compares our work with that of others. And, Section 7 presents our conclusions. 2 The Benchmark Suite Table 1 lists all the codes in our sparse irregular benchmark suite[1]. We chose them for the same reason programs were chosen for the collection of HPF 2 motivating applications: they include parallel idioms important to full scale Table 2. Loop Patterns Indirectly Accessed Array Others Right Offset Sparse Consecutively Premature Array Benchmark hand Histogram ....

Rafael Asenjo, Eladio Gutierrez, Yuan Lin, David Padua, Bill Pottenger, and Emilio Zapata. On the Automatic Parallelization of Sparse and Irregular Fortran Codes. Technical Report 1512, Univ. of Illinois at Urbana-Champaign, CSRD, Dec 1996

....basic operation on this code is the computation of physical magnitudes (such as forces) corresponding to the nodes described by a mesh. The magnitudes are computed over the mesh edges, each one de ned by two nodes. Therefore two subscript arrays are needed to compute the magnitudes of each edge [1, 8]. This reduction loop is interesting from the parallelization point of view because it contains subscripted reads and writes. In order to avoid side e ects di erent from the irregular reductions, all experiments presented in this section only consider one of the reduction loops included in the ....

....in this section only consider one of the reduction loops included in the EULER code. Speci cally, the loop shown in Fig. 9, which corresponds to a single static loop with reductions using two subscript arrays. From the HPF 2 motivating applications suite we have extracted also the code NBFC [1, 3], which carries out a molecular dynamics simulation. Speci cally, NBFC computes electrostatic interactions, that is, non bonded forces, between particles. In order to reduce the oating point count, a precomputed list of neighbor particles is used, which causes the existence of an irregular ....

[Article contains additional citation context not shown here]

R. Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottengerg, and E. Zapata. On the automatic parallelization of sparse and irregular Fortran codes. Technical Report TR-1512, University for Illinois at Urbana-Champaign. Center for Supercomputing R&D, December 1996.

....basic operation on this code is the computation of physical magnitudes (such as forces) corresponding to the nodes described by a mesh. The magnitudes are computed over the mesh edges, each one defined by two nodes. Therefore two subscript arrays are needed to compute the magnitudes of each edge [1, 10]. This reduction loop is interesting from the parallelization point of view because it contains subscripted reads and writes. In order to avoid side effects different from the irregular reductions, all experiments presented in this section only consider one of the reduction loops included in the ....

....in this section only consider one of the reduction loops included in the EULER code. Specifically, the loop shown in Fig. 11, which corresponds to a single static loop with reductions using two subscript arrays. From the HPF 2 motivating applications suite we have extracted also the code NBFC [1, 3], which carries out a molecular dynamics simulation. Specifically, NBFC computes electrostatic interactions, that is, nonbonded forces, between particles. In order to reduce real A(ADim) integer f1(fDim) f2(fDim) f3(fDim) integer init(NumThreads,NumThreads) integer ....

[Article contains additional citation context not shown here]

R. Asenjo, E. Guti'errez, Y. Lin, D. Padua, B. Pottengerg and E. Zapata, On the Automatic Parallelization of Sparse and Irregular Fortran Codes, Technical Report 1512, University for Illinois at UrbanaChampaign, Center for Supercomputing R&D., December 1996.

....would be desirable to have a compiler capable of selecting the appropriate data distribution with no intervention from the programmer. However, this is a very tough task for numerical irregular problems. Currently the problem is open and unsolved, except for some simple classes of irregular codes [10,9,2]. The compilation task can be simplified if the programmer is allowed to be involved through the use of compiler directives. These directives may help the compiler in identifying the class of irregular application, that is the role of the main data structures in the code, as well as express ....

R.Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottengerg and E. Zapata, On the Automatic Parallelization of Sparse and Irregular Fortran Codes, Technical Report 1512 , (Univ. for Illinois at Urbana-Champaign, CSRD, December 1996).

....as in the (LocalWrite) technique [5] although it is not reported as a clear good alternative to parallelize irregular reductions) In a shared memory context, academic parallelizers like Polaris [2] and SUIF [4] recognize and parallelize irregular reductions. A number of techniques are available [7, 1]: critical sections, data affinity, privatized buffer (SUIF) array expansion (Polaris) and reduction table. The most efficient techniques, privatized This work was supported by the Ministry of Education and Science (CICYT) of Spain (TIC96 1125 C03) real A(1:ADim) integer f(1:fDim) do i = ....

R. Asenjo, E. Gutierrez, Y. Lin, D. Padua, B. Pottengerg and E. Zapata, On the Automatic Parallelization of Sparse and Irregular Fortran Codes, TR--1512, Univ. of Illinois at Urbana-Champaign, Ctr. for Supercomputing R&D., Dec. 1996.

No context found.

Rafael Asenjo, Eladio Gutierrez, Yuan Lin, David Padua, Bill Pottenger, and Emilio Zapata. On the Automatic Parallelization of Sparse and Irregular Fortran Codes. Technical Report 1512, Univ. of Illinois at Urbana-Champaign, Center for Supercomputing Res. & Dev., December 1996.

No context found.

) Asenjo, R., Gutierrez, E., Lin, Y., Padua, D., Pottengerg, B. and Zapata, E.L.: On the Automatic Parallelization of Sparse and irregular Fortran codes, Technical Report 1512, University of Illinois at Urbana-Champaign, CSRD (Dec. 1996).

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