Reinhard von Hanxleden. Compiler Support for Machine-Independent Parallelization of Irregular Problems. PhD thesis, Rice University, December 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....to place communication and inspection are loop based, using data dependence analysis, and that restricts the optimizations possible in the presence of conditionals. These simplifications are key to reduce the complexity of the compiler and it is unclear whether more complex analysis techniques [70] would pay off in terms of performance of real applications. 4.3 Code Normalizations The normalization of loop bounds and array subscripts is crucial to reduce the complexity of the rest of the compiler. The implementation of this pass was greatly simplified by the use of Mathematica. A ....

....ffl When multi dimensional arrays are only accessed irregularly in some dimensions but not others. ffl When most of the array references contained in the inspector slice of an irregular reference are regular. Previous work focussed on data flow frameworks to either place irregular communication [70] or generate inspectors with multiple levels of indirection [18] Unfortunately, these studies treat communication and inspection placement as two independent problems, and the issues of a nontrivial integration are not discussed. Our approach is to consider these problems together but in a more ....

[Article contains additional citation context not shown here]

Reinhard von Hanxleden. Compiler Support for Machine-Independent Parallelization of Irregular Problems. PhD thesis, Rice University, December 1994.

....accesses to shared arrays, in order to reduce page traffic in a SVM architecture. Other techniques have been already proposed for DMPCs which do not support a SVM. PARTI [8] is one of the most advanced projects in resolving such problems. It has been grafted to several compilers such as FORTRAN D [6] and Vienna Fortran[4] Our proposal is an adaptation of the inspector executor scheme, introduced in PARTI. It has been implemented within the Fortran S compiler [1] which is targeted to several SVM DMPCs. Comparing to other approaches, based on message passing, our techniques requires less ....

R. Von Hanxleden. Compiler Support for MachineIndependent Parallelization of Irregular Problems. PhD thesis, Rice University, Houston, TX, December 1994.

....what the optimal compiler could produce and are therefore interesting for performance comparisons. The SPLASH benchmark suite [SWG92] for distributed memory computers contains many reallife applications, some of which can be converted for our system and used as examples. For the FORTRAN D system, [CvHMS92, vHKS93, vH92] examine techniques to facilitate the parallel execution of FORTRAN programs. They concentrate on efficient communication with data prefetch and redundancy elimination and gather the necessary data with inspector executor loops. There is also a runtime system that performs spatial ....

Reinhard von Hanxleden. Compiler support for machine independent parallelization of irregular problems. Thesis Proposal, November 1992.

.... Compiler technology that exploits such research by transforming sequentially consistent programs into programs which dynamically select the appropriate consistency model is very much in its infancy [10] VSM optimisations which are similar to those described in this report have been presented in [13, 8, 5]. Previous work has not presented detailed application studies, or has provided experimental results for implementations of VSM optimisations that are believed to be unprotected 1 . The work presented in this report describes how protected implementations of local invalidate and local exclusive ....

....the write. This cannot occur until all nodes which require a read copy have executed their read, so the local exclusive must be placed after the first barrier and before the write. We therefore remove unnecessary invalidation traffic and node stalls, see Figure 1b. A similar approach is used in [13] when implementing their EM3D code using Tempest, a coherence interface, on Typhoon, a simulated VSM architecture. 2.3.2 Iterative Solver The code described in this Section implements a simple matrix iteration. The communication pattern represents the communication pattern found in iterative ....

[Article contains additional citation context not shown here]

Reinhard von Hanxleden. Compiler Support for Machine-Independent Parallelization of Irregular Problems. PhD thesis, Rice University, Houston, TX, December 1994.

....is developed by us in analogy to the latter method. Data dependencies between array references have to be taken into account. When expensive operations like remapping are in question, smaller parts of an array could be useful: data flow analysis is applied for array sections by von Hanxleden [vH94]. Detailed consideration of array sections is subject of further work. It remains to be checked whether a refinement by using intermediate representation is worth the effort. If a low intermediate representation is used, prefetching, the preloading of data from memory into cache ahead of time ....

R. von Hanxleden. Compiler Support for Machine-Independent Parallelization of Irregular Problems. PhD thesis, Rice University, Texas, 1994.

....Banerjee [GuB92] claim an average error margin of less than 10 for their techniques they are generally only applicable for highly regular parallel programs. Unfortunately, there are a host of problems, including many of the so called Grand Challenge problems, that are irregular 1 in nature [Han92]. Such unpredictability exacerbates the difficulty of providing reasonable computation and communication timing estimates. To illustrate the potential impact of using accurate versus intrusively obtained or estimated timing information in task scheduling, consider the functional abstraction of a ....

R. von Hanxleden, "Compiler Support for Machine Independent Parallelization of Irregular Problems," CRPC-TR92301-S, Center for Research on Parallel Computation, Rice University, November 1992.

....is used to capture the input dependent nature of irregular problems, and is similar in spirit to our data structures used for the adaptive FMM algorithm. The Fortran D compiler [1] has also been extended in an attempt to deal with irregular problems, in the thesis of Reinhard von Hanxleden [30]. The Inspector Executor model is used for generating input dependent data mapping. There are many other heuristics for task graph scheduling in the literature. PYRROS belongs to the class of multistage methods, which also includes the Sarkar s algorithm [47] among many others. Some other methods ....

Reinhard von Hanxleden, Compiler support for machine-independent parallelization of irregular problems, Ph.D. Thesis, Rice University, Dec. 1994.

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