T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Boston, USA, ISBN 0-79239708 -8, March 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....to benefit from these efforts . Privatization was not supported in either the Fortran D [12] or Vienna Fortran [16] but has since been added to the HPF language definition [9] Whenever situations arose that could have been handled by privatization, techniques such as global forward substitution [17] were commonly applied instead. More recently, a prototype HPF compiler being developed as part of the SP2 project [18] has addressed this issue in a different manner. In this project, scalars are assumed to be replicated by default, but for each definition of the scalar that can be privatized, ....

T. Fahringer, Automatic Performance Prediction for Parallel Programs on Massively Parallel Computers. PhD thesis, University of Vienna, Vienna, Austria, Sept. 1993. TR93-3.

.... VFC compiler [2] translates HPF programs into message passing Fortran90 programs based on MPI) We are currently also investigating the usefulness of SCALA for performance analysis of distributed JAVA programs [16] Moreover, we are in the process to integrate SCALA with performance prediction [14] and symbolic analysis techniques [15] to examine the scaling behavior [42] of distributed and parallel programs. Finally, we also work on fully automatizing the process of performance analysis [17] This means that performance information is interpreted by SCALA which then automatically detects ....

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Boston, USA, ISBN 0-79239708 -8, March 1996.

....et al. [13] describe their experiences with modeling two message passing parallel programs. Simulation using the application models allows prediction of application performance and resource usage on large scale systems and problems as well as identification of performance bottlenecks. Fahringer [5] introduces an approach for predicting performance parameters of parallel programs based on compiler analysis and machine specific parameters. His experiments demonstrate accurate performance estimates as well as the ability to guide both programmers and compilers in optimizing parallel programs. ....

FAHRINGER, T. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Norwell, Massachusetts, 1996.

....important slowdown. The aim of tools based on high level language is to allow adapted accuracy in designing parallel applications with a minimum slowdown. Mainly, this tool category has the ability to calculate application performance on various types of architecture from an execution trace. P3T [Fah96] Dimemas [GCL97] Patop [WOKH96] and ChronosMix [BST99] belong to this category. The P3T tool is part of the Vienna Fortran Compilation System and its aim is to evaluate and classify parallel strategies. P3T helps the compiler to nd the appropriate automatic parallelization of the sequential ....

T. Fahringer. Automatic Performance Prediction of Parallel Programs. ISBN 0-7923-9708-8. Kluwer Academic Publishers, Boston, USA, March 1996.

....TIC 95 0868 problems of the application and shows them to the application programmer, together with source code references of the problem found, and indications on how to overcome the problem. The main difference between the KAPPA PI tool and the existing automatic performance analyisis tools [2] [3] 4] is that the code of the analysed application is checked to propose alternatives for a new behaviour. Analysis first considers the study of the trace file in order to locate the most important performance problems occurring at the execution. Once those problematic execution intervals have ....

Fahringer, T. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers. 1996.

.... tting as investigated in [3] The problem of choosing a data partitioning and distribution to achieve optimal performance is NP complete [4] so we are more interested in user guided performance evaluation tools for re ning parallel applications than in automatic performance prediction (e.g. [5]) Since our model is intended for use in the context of archetype based application development, it di ers from e orts to do performance measurement for compiler optimization (e.g. 6] and from e orts to estimate performance statically to automate load balancing (e.g. 7] Our techniques t ....

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, 1996.

....notions of volume, namely the volumes w.r.t. 1. the well known Lebesgue measure (cf. definition 2.21) 2. and a certain integer counting measure (cf. definition 2. 23) Methods that can be used in order to compute the first kind of volume of a single convex polyhedron are presented, for instance, in [Fah93], HKSF94] Ber95] Fah96] In contrast to this we won t compute a number that represents the volume of a single polyhedron but a function depending on some parameters that yields the volume of a polyhedron that is member of a parametrised family of polyhedra. The computation of the volume of a ....

T. Fahringer. Automatic Performance Prediction for Parallel Programs on Massively Parallel Computers. PhD thesis, Institute of Software Technology and Parallel Systems,University of Vienna, October 1993.

....when either the number of spectral bands or the image resolution is increased. Our experiments show that even for small numbers of spectral bands (8 12) the sequential overhead does not exceed 10 . Figure 6. Performance Chart It can be shown, using techniques developed by Foster et.al. [11, 20], that the performance of the spectralscreening PCT can be described analytically with the following predictive model: 0 1 2 3 4 5 6 7 8 9 0 1 2 3 LOG(N) ideal measured 8 2 2 7 3 6 2 5 2 4 3 2 2 2 1 ) 1 ( C p m n C n C p n C p s n C p sn C p s p C p snm C ....

T. Fahringer, Automatic Performance Prediction of Parallel Programs, Kluwer Academic Publishers, Boston/London, 1996.

....research effort to build modeling tools because such tools are imperative to derive efficient implementations. The significant work includes the work related to the Fortran D compiler [4] the Paradigm compiler [5] the Suif compiler [2] the Fx compiler [34, 33] and the Vienna Fortran Compiler [11, 12]. Other approaches include the use of petri nets [13] queuing networks, and Markov chains [35] The Fortran D compiler contains an interactive tool that allows the programmer to select regions of the sequential input program. The tool responds with a data decomposition scheme and diagnostic ....

T. Fahringer. Automatic Performance Prediction for Parallel Programs on Massively Parallel Computers. PhD thesis, University of Vienna, 1993.

....processors, and build a symbolic model that represents the predicted execution time as a function of those parameters. The derivation of this model, however, required statistical methods and several experimental runs of the program with different problem sizes and numbers of processors. Fahringer [22] designed a performance prediction tool named PPPT (Parameter based Performance Prediction Tool) which analyzes a set of parameters that characterize the behavior of a parallel program, including work distribution, amount of communication and data locality. The tool correlates statically computed ....

....studies have been conducted in this direction, trying to integrate automatic prediction and compilation tools. However, that effort has not yet reached an ideal stage. Some methods derive a prediction for a specific combination of number of processors (P ) and problem size (N ) like in [1] [22] and [50] Others provide a symbolic model that can be evaluated at desired combinations of N and P , but either have a very limited application domain, as in [64] or require several executions of the program for model calibration, as in [15] There has been no proposed method, so far, that ....

Thomas Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Norwell, Massachusetts, 1996.

....estimated, then a kernel selection algorithm (like the one presented in 4.2) is applied on the whole set of fragments. If the data set does not fully describe the complexity of the source program then some kernels may be forgotten. In addition, whether the kernel detection is performed at runtime [Fah93] or by program complexity computation, at least one execution of the source program is needed 3 . This is costly and and may be even impossible. 3 The complexity of the fragments may indeed be a function of some quantities only measurable at run time like the probability that a structured if ....

T. Fahringer. Automatic Performance Prediction for Parallel Programs on Massively Parallel Computers. PhD thesis, Dept of Comp. Science., University of Vienna, November 1993.

.... of choosing the data partitioning and distribution to achieve the optimal performance is NP complete, we are more interested in user guided performance evaluation tools for the refinement of parallel applications than in automatic performance prediction (for example, Fahringer s work with P 3 T [Fah96a] Since our model is likely to be used to supplement a programmer s efforts to develop applications using archetypes, it differs from efforts to do performance measurement for compiler optimization (as Clement and Quinn do with C on multicomputers [CQ93] and it differs from efforts to ....

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, 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 ....

T. Fahringer. Automatic Performance Prediction for Parallel Programs on Massively Parallel Computers. PhD thesis, University of Vienna, 1993.

No context found.

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Boston, USA, ISBN 0-79239708 -8, March 1996.

No context found.

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Boston, USA, ISBN 0-7923-9708-8, March 1996.

No context found.

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Boston, USA, ISBN 0-7923-9708-8, March 1996.

No context found.

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Boston, USA, ISBN 0-79239708 -8, March 1996.

No context found.

T. Fahringer. Automatic Performance Prediction of Parallel Programs . Kluwer Academic Publishers, Boston, USA, ISBN 0-7923-9708-8, March 1996.

....remarks are given in Section 8. 2 Preliminaries The following notations and definitions are used in the remainder of this paper: ffl Our symbolic analysis has been implemented and is currently being integrated with VFCS [2] a HighPerformance Fortran style parallelizing compiler and P T [8, 12], and with a performance estimator for data parallel programs on distributed memory parallel architectures. The VFCS paralleliziation strategy is based on data decomposition in conjunction with the single program, multiple data programming model. With this method, each array is partitioned and ....

....serialize loop L5 as they fail to evaluate non linear array subscript expressions. 7 4 Count Solutions to a System of Constraints Counting the number of integer solutions to a set of constraints has been shown to be a key issue in performance analysis of parallel programs. Numerous applications [8, 21, 14] include: estimating statement execution counts, branching probabilities, work distribution, number of data transfers and cache misses. Even compiler analysis can be supported, for instance, by detecting and eliminating dead code such as loops that never iterate (zero trip loops) In what ....

[Article contains additional citation context not shown here]

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Boston, USA, ISBN 0-7923-9708-8, March 1996.

....and control flow such as statement execution and average loop iteration counts. We use a single sequential profile run (cf. Section 4) of the original input program ignoring all explicit parallel language constructs on a single processor. For this purpose, we developed the Weight Finder ([3]) which is an advanced and highly optimized automatic profiler for Fortran programs. Profile runtime, memory requirements, and perturbation on the program s behavior is reduced by minimizing instrumentation overhead. Instrumentation is done by automatically inserting time measurement and event ....

....and eliminating redundant instrumentation, hoisting instrumentation code out of loops, and computing profile information instead of counting events whenever possible. Selective instrumentation with respect to profile data and code segments of interest further reduces the profile overhead. In [3] we have shown that large portions of the profile data can be successfully adapted for most important program transformations without redoing the profile run. However, for changing the problem size, we currently have to repeat the profile run. We are investigating scalability methods such as those ....

[Article contains additional citation context not shown here]

T. Fahringer. Automatic Performance Prediction for Parallel Programs on Massively Parallel Computers. PhD thesis, University of Vienna, Institute forSoftware Technology and Parallel Systems, October 1993.

....models are commonly used to assume a more or less virtual and often unrealistic application behavior. Moreover, very few performance estimators actually consider code transformations and optimizations applied by a compiler. In this paper we introduce P 3 T , the successor tool of P 3 T [22, 15, 16], which models programs, code transformations, and parallel and distributed architectures. The input programs of P 3 T are written in High Performance Fortran [27, 1] which represents the de facto standard of high level data parallel programming. Moreover, P 3 T analyzes Fortran90 message ....

.... statement S is defined by wd(S) 1 owork(S) 2 v u u t 1 jP A j X p2P A Gamma work(S; p) Gamma owork(S) Delta 2 The above formula is the standard deviation (oe) divided by the arithmetic mean (owork(S) which is known as the variation coefficient in statistics [4] In [16] we have presented a proof for the lower and upper bound of wd(S) with the following result: 0 wd(S) jP A j Gamma 1. Best case and worst case work distribution are, respectively, given by wd(S) 0 and wd(S) jP A j Gamma 1. Based on Definition 4.2, a work distribution goodness function ....

[Article contains additional citation context not shown here]

T. Fahringer. Automatic Performance Prediction of Parallel Programs. Kluwer Academic Publishers, Boston, USA, ISBN 0-79239708 -8, March 1996.

No context found.

T. Fahringer, Automatic performance prediction of parallel programs, in: Automatic Performance Prediction of Parallel Programs, Boston, USA, March 1996, Kluwer Academic Publishers, Dordrecht, ISBN 0-7923-9708-8.

No context found.

T. Fahringer, "Automatic performance prediction for parallel programs on massively parallel computers," Ph.D. dissertation, University of Vienna, Austria, Sept. 1993, TR93-3.

No context found.

T. Fahringer. Automatic Performance Prediction for Parallel Programs on Massively Parallel Computers. PhD thesis, University of Vienna, Vienna, Austria, Sept. 1993. TR93-3.

No context found.

Fahringer T., Automatic Performance Prediction of Parallel Programs, Kluwer Academic Publishers, Boston/London, 1996.

First 50 documents  Next 50

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