S. Ramaswamy, "Simultaneous exploitation of task and data parallelism in regular scientific applications," Ph.D. dissertation, Department of Electrical and Computer Engineering, University of Illinois, Urbana, IL, Jan. 1996, CRHC-96-03/UILU-ENG-96-2203.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....or any other descriptor that includes all processors executing the concurrent tasks. Moreover, Fortran M and Fx do not allow recursive specifications of task parallelism. An exploitation of task and data parallelism in the context of a parallelizing compiler can be found in the Paradigm project [11, 15]. The Paradigm compiler provides a framework that expresses task parallelism by a macro data flow graph which has been derived from the hierarchical task graphs used in the Parafrase compiler [3] A major difference between the Paradigm and the Tlib approach is also that Paradigm expects a ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications. PhD thesis, University of Illinois at Urbana-Champaign, 1996.

....groups. Previous work on supporting multi level parallelism focused on providing some kind of coordination support to allow the interaction of a set of program modules (task parallelism) in the framework of HPF [11] parallel programs (data parallelism) for distributed memory architectures [5,6,8,26]. Recently, KAI (Kuck and Associates, Inc. has made proposals to OpenMP to support multi level parallelism through the WorkQueue mechanism [12] in which work can be created dynamically, even recursively, and put into queues. Within the WorkQueue model, nested queuing permits a hierarchy of ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Computations. Ph.D. Thesis. University of Illinois at Urbana-Champaign, 1996.

....the interaction of a set of program modules in the framework of data parallel programs for distributed memory architectures. For example, 3] proposed a library based approach that provides a set of functions for coupling multiple HPF tasks to form task parallel computations. Other alternatives [2, 6, 10] proposed a small set of Fortran directives to integrate task and data parallelism also in an HPF framework. Our group has proposed extensions to OpenMP in order to allow an efficient exploitation of nested parallelism. The extensions offer the concept of thread groups [5] This concept is similar ....

....to having the appropriate support at the language level to express generic task graphs including pipelined schemes, good scheduling techniques are required to map the different levels of parallelism onto the available processors. A large number of heuristics have been proposed in the literature [12, 9, 10] for mixed task and data parallel scheduling. The proposal in this paper includes a set of new directives and clauses for the emerging indutrial standard OpenMP to specify generic task graphs and an associated processor mapping. The paper does not assume any particular scheduling heuristic. The ....

[Article contains additional citation context not shown here]

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Computations. Ph.D. Thesis, University of Illinois at Urbana--Champaign, 1996.

....programs for distributed memory architectures. Some of them combine the use 1 of two programming models and interfaces. For example, 3] proposes a library based approach that provides a set of functions for coupling multiple HPF tasks to form task parallel computations. Other alternatives [2, 5, 16] proposed a small set of Fortran directives to integrate task and data parallelism parallelism also in an HPF framework. The Illinois Intel Multithreading library [4] targets shared memory systems. It also supports multiple levels of general (unstructured) parallelism. Application tasks are ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Computations. PhD Thesis. University of Illinois at Urbana-- Champaign, 1996.

....interaction of a set of program modules in the framework of data parallel programs for distributed memory architectures. For instance, 9] propose a library based approach that provides a set of functions for coupling multiple HPF tasks to form task parallel computations. The Fx [10] and PARADIGM [11] projects propose extensions to integrate task and data parallelism in an HPF environment. The use of task parallelism is proposed to improve the performance when data parallelism is not enough. Our thread package implementation (NthLib [12] is built based on the Nano Threads Programming Model ....

Ramaswamy, S., "Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Computations", Ph.D. Thesis, Univ. of Illinois at Urbana-Champaign, 1996.

....techniques, and parallelizing compilers. Related work on programming paradigms includes the BSP model [36] the logP model [4] and the logGP model [1] see [19] for an overview of further models. Related work on performance prediction includes the studies in the context of parallelizing compilers [9, 21] as well as other prediction models [7, 5] Several research groups working on parallelizing compilers have included support to combine task and data parallelism. Fortran M [16, 11] allows the creation of processes which can communicate with each other by predefined channels and which can be ....

....into consideration and the parameter of which are determined by separate runtime tests for each application. There is no verification of the predicted runtimes in [17] An exploitation of task and data parallelism in the context of a parallelizing compiler can be found in the Paradigm compiler [20, 21, 32]. The Paradigm compiler provides a framework that expresses task parallelism by a macro dataflow graph which is similar to a graphical representation of the upper half of a module specification. There are two main differences between the Paradigm and the TwoL approach. First, Paradigm expects as ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications. PhD thesis, University of Illinois at UrbanaChampaign, 1996.

.... A good overview of related work on programming paradigms can be found in [11] Related work on computation models includes the BSP model [24] the logP model [4] and the logGP model [1] Related work on performance prediction includes investigations in the context of parallelizing compilers [7, 12] as well as other prediction models [6] In the following, we compare the TwoL approach to similar approaches in the area of parallelizing compilers. Several research groups working on parallelizing compilers have included support to combine task and data parallelism. Fortran M [9, 8] allows the ....

....into consideration and the parameter of which are determined by separate runtime tests for each application. There is no verification of the predicted runtimes in [20] An exploitation of task and data parallelism in the context of a parallelizing compiler can be found in the Paradigm compiler [2, 12, 13]. The Paradigm compiler provides a framework that expresses task parallelism by a macro dataflow graph which is similar to a graphical representation of a module specification in TwoL. There are two main differences between the Paradigm and the TwoL approach. First, Paradigm expects as input a ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications. PhD thesis, University of Illinois at Urbana-Champaign, 1996.

....lead to an a priori estimation of the prospective gain of a parallel implementation. Thus, the programmer can decide on the benefits of a parallel implementation before actually performing it. An important application area for analytical performance prediction mechanisms are compilers for DMMs [21, 4, 22, 23]. These compilers assist the programmer in generating an efficient parallel implementation. To make suggestions for the design of a parallel implementation, e.g. a selection of a data distribution, the compiler has to have access to a powerful performance prediction tool to estimate the effects ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications. PhD thesis, University of Illinois at Urbana-Champaign, 1996.

....the program and representing the body of each loop by an acyclic control flow graph. Algorithms for the dynamic scheduling of task graphs are presented in [7, 6] A similar graph, the hierarchical macro dataflow graph (MDG) is used in the Paradigm compiler to represent data and task parallelism [24, 22]. Nodes in the MDG correspond to basic parallel tasks or loop constructs, edges in the MDG correspond to precedence constraints that exist between tasks. The nodes and edges of the MDG are weighted with processing and data transfer costs both of which depend on the number of processors used for ....

....MDG correspond to basic parallel tasks or loop constructs, edges in the MDG correspond to precedence constraints that exist between tasks. The nodes and edges of the MDG are weighted with processing and data transfer costs both of which depend on the number of processors used for the execution. [22] describes scheduling and allocation algorithms for MDGs where allocation decides on the number of processors to use for each node and scheduling decides on a scheme of execution for the allocated nodes. The goal is to select a strategy that minimizes the execution time of the MDG. 23] considers ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications. PhD thesis, University of Illinois at Urbana-Champaign, 1996.

....because it is a very difficult combinatorial optimization problem [5] 6] There are researches on parallelizing compilers that optimize execution time considering both computation and communication. In particular, PARADIGM compiler [7] optimizes execution time by using convex programming [8][9]. However, the estimation model in PARADIGM is still too much simplified and the derived solution is suboptimal, because optimization problem is relaxed to continuous domain to make the problem easy to solve. Contrary, a combinatorial optimization problem is solved in our research to derive the ....

S. Ramaswamy, Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications, Ph.D. Thesis, Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL (1996).

....8 Introduction stream. The PTRAN compiler [Sar91] exploits parallelism among loop iterations using the PDG, but does not exploit parallelism at other granularities. The thesis exploits parallelism across the full range of granularities for one or more instruction streams. Ramaswamy [RB93, Ram91] produced some leading research in exploiting both control and data parallelism for distributed shared memory machines. His work focuses on overlapping independent data parallel loops and models the execution time of each task in terms of Amdahl s law. That approach is less general than the one ....

....executes a loop in parallel across all processors or executes it serially. Other PDG related work [GS89, AJLS92] and HTG related work [Wal92] exploits parallelism at different granularities, but only for a single instruction stream. Work based on the HTG and the Macro Dataflow Graph (MDG) RB93, Ram91] is most similar to the techniques presented here. Some of the most important differences are that work does not consider data dependence latencies among overlapped constructs, and it assumes that the computation time of each construct can be modeled by Amdahl s law, thus implicitly assuming the ....

Shankar Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications. PhD thesis, University of Illinois, 1991.

....execution schemes of the DIIRK method on the IBM SP2 (left) and Intel Paragon (right) dense input system. 6 Comparison with Related Work Work related to the TwoL approach includes studies on new parallel programming paradigms [18] computation models [1, 7, 31] performance prediction techniques [9, 17, 19], and parallelizing compilers. In the following, we compare our approach to similar approaches in the area of parallelizing compilers. Several research groups working on parallelizing compilers have included support to combine task and data parallelism. Fortran M [11, 12] allows the creation of ....

....is similar to 7 CONCLUSIONS AND FUTURE RESEARCH 18 the task parallelism model of Fx has recently been added to High Performance Fortran [16] as an approved extension. An exploitation of task and data parallelism in the context of a parallelizing compiler can be found in the Paradigm compiler [3, 19, 21]. The Paradigm compiler provides a framework that expresses task parallelism by a macro data flow graph which has been derived from the hierarchical task graphs used in the Parafrase compiler [4] Nodes in the macro data flow graph correspond to basic parallel tasks or loop constructs, edges ....

[Article contains additional citation context not shown here]

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications. PhD thesis, University of Illinois at Urbana-Champaign, 1996.

....The Fx project [GrOS94] focused on proposing a small set of Fortran directives to integrate task and data parallelism parallelism also in an HPF framework; the framework targets a class of applications that process continuous streams of data sets. In a similar way, the PARADIGM project [Rama96] also proposes the use of task parallelism to control the degree of data parallelism in individual tasks; they extend the model proposed in the Fx project in order to target a more general class of applications and to perform allocation and scheduling of processors to HPF data parallel routines ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Computations. PhD Thesis. University of Illinois at Urbana--Champaign, 1996.

.... Communication operations in the PBM are translated into their MPI counterparts (not shown in the Figure) 5 Comparison with Related Work Work related to the TwoL approach includes studies on new parallel programming paradigms [15] computation models [27, 7, 1] performance prediction techniques [16, 8, 14], and parallelizing compilers. In the following, we compare our approach to similar approaches in the area of parallelizing compilers. Several research groups working on parallelizing compilers have included support to combine task and data parallelism. Fortran M [10, 9] allows the creation of ....

....tests for each application. A model that is similar to the task parallelism model of Fx has recently been added to High Performance Fortran [13] as an approved extension. An exploitation of task and data parallelism in the context of a parallelizing compiler can be found in the Paradigm compiler [3, 16, 18]. The Paradigm compiler provides a framework that expresses task parallelism by a macro dataflow graph which has been derived from the hierarchical task graphs used in the Parafrase compiler [4] Nodes in the macro dataflow graph correspond to basic parallel tasks or loop constructs, edges ....

[Article contains additional citation context not shown here]

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Applications. PhD thesis, University of Illinois at Urbana-Champaign, 1996.

....The Fx project [GrOS94] focused on proposing a small set of Fortran directives to integrate task and data parallelism parallelism also in an HPF framework; the framework targets a class of applications that process continuous streams of data sets. In a similar way, the PARADIGM project [Rama96] also proposes the use of task parallelism to control the degree of data parallelism in individual tasks; they extend the model proposed in the Fx project in order to target a more general class of applications and to perform allocation and scheduling of processors to HPF data parallel routines ....

S. Ramaswamy. Simultaneous Exploitation of Task and Data Parallelism in Regular Scientific Computations. PhD Thesis. University of Illinois at Urbana--Champaign, 1996.

No context found.

S. Ramaswamy, "Simultaneous exploitation of task and data parallelism in regular scientific applications," Ph.D. dissertation, Department of Electrical and Computer Engineering, University of Illinois, Urbana, IL, Jan. 1996, CRHC-96-03/UILU-ENG-96-2203.

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