F. McMahon. The Livermore Fortran Kernels: A computer test of the numerical performance range. Technical Report UCRL-53745, Lawrence Livermore National Laboratory, December 1986.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Algebra Subroutines) Law79, Don79] These vector and matrix computations occur frequently in scientific applications, thus they have been collected into a set of library routines that are highly optimized for various host architectures. Hydro and tridiag are the first and fifth Livermore Loops [McM86], a set of kernels culled from important scientific computations. The former is a fragment of a hydrodynamics computation, and the latter is a tridiagonal elimination computation. Although the computations differ, their access patterns are identical, thus results for these benchmarks are ....

McMahon, F.H., "The Livermore Fortran Kernels: A Computer Test of the Numerical Performance Range", Lawrence Livermore National Laboratory, UCRL-53745, December 1986.

....is shown in Figure 5.13, and Figures 5.14 and 5.15 show the performance results. The source codes generated by PARADIGM for both versions are shown in Appendix A. 5. 4 Two Dimensional Explicit Hydrodynamics Kernel The two dimensional explicit hydrodynamics kernel from the set of Livermore kernels [26] is the main set of computation loops extracted from a hydrodynamics application. The computation is naturally expressed using the data parallel FORALL construct. The original source code is shown in Figure 5.16, and the source code generated by PARADIGM is shown in Appendix A. Figures 5.17 and ....

F. McMahon, "The Livermore Fortran Kernels: A computer test of the numerical performance range", Tech. Rep. UCRL-53745, Lawrence Livermore National Laboratory, Livermore, CA, Dec. 1986.

....7 block, Jacobi 500 Theta500 1000 Theta1000 block, block 4. EVALUATION OF OPTIMIZATIONS A group of small scientific program kernels are used to examine the performance of the presented communication optimizations. The selected program fragments include ffl ADI Integration (ADI, kernel 8) [25] ffl 2 D Explicit Hydrodynamics (EXPL, kernel 18) 25] ffl Jacobi s Iterative Method Three other programs which exhibit cross iteration dependencies are selected to examine the pipelining optimization: ffl Implicit Hydrodynamics (IMPL, kernel 23) 25] ffl Successive Over Relaxation Iterative ....

....block 4. EVALUATION OF OPTIMIZATIONS A group of small scientific program kernels are used to examine the performance of the presented communication optimizations. The selected program fragments include ffl ADI Integration (ADI, kernel 8) 25] ffl 2 D Explicit Hydrodynamics (EXPL, kernel 18) [25] ffl Jacobi s Iterative Method Three other programs which exhibit cross iteration dependencies are selected to examine the pipelining optimization: ffl Implicit Hydrodynamics (IMPL, kernel 23) 25] ffl Successive Over Relaxation Iterative Method (SOR) ffl Block Lower Triangular Solver (BLTS) ....

[Article contains additional citation context not shown here]

F. McMahon, "The Livermore Fortran Kernels: A computer test of the numericalperformance range," Tech. Rep. UCRL-53745, Lawrence Livermore National Laboratory, 1986.

....at the very same time, in the order specified by the priority function. 4 Experimental results 4. 1 The experimental setup The experimental input to the research scheduler was obtained from the Perfect Club benchmark suite [6] the Spcc benchmarks [43] and the Livermore Fortran Kernels (LFK) [27] using the Fortran77 compiler for the Cydra 5. The Cydra 5 compiler examines every innermost loop as a potential candidate for modulo scheduIing. Candidate loops are rejected if they are not DO loops, if they can exit early, if they contain procedure calls, or if they contain more than 30 basic ....

McMahon, F.H. The Livermore Fortran kernels: a computer test of the numerical performance range. Technical Report UCRL-53745. Lawrence Livermore National Laboratory. Livermore, California, 1986.

....Linear Algebra Subroutines) Law79, Don79] These vector and matrix computations occur frequently in scientific computations, thus they have been collected into libraries of highly optimized routines for various host architectures. Hydro and tridiag are the first and fifth Livermore Loops [McM86], a set of kernels culled from important scientific computations. The former is a fragment of a hydrodynamics computation, and the latter is a tridiagonal elimination computation. Vaxpy is a vector axpy computation that occurs in matrix vector multiplication by diagonals; this algorithm is ....

McMahon, F.H., "The Livermore Fortran Kernels: A Computer Test of the Numerical Performance Range", Lawrence Livermore National Laboratory, UCRL-53745, December 1986.

....of knowing exactly how many machine instructions are required to implement a given section of code. 4 Results In order to demonstrate the effectiveness of our instruction partitioning mechanism, we conducted a number of experiments using kernels from Linpack [3] Livermore Loop Fortran Kernels [14] (LLFK) and multimedia benchmarks from suites such as MediaBench [12] We used a simple RISC based simulator to run the compiled examples on a variety of cache architectures. As a simple example of typical results, we present a Fast Fourier Transform or FFT kernel which is used in many ....

F.H. McMahon. The Livermore Fortran Kernels: A Computer Test Of The Numerical Performance Range. Lawrence Livermore National Laboratory, Livermore, California, December 1986.

....results than the actual performance. 4.1.2.2 Kernel Benchmarks and Compact Application Benchmarks To avoid the problems of synthetic benchmarks, some kernel benchmarks have been proposed. Many benchmarks fall in this category. Some examples are the Linpack benchmark [9] the Livermore benchmark [21], and part of the MITRE benchmark [13] However, these kernel benchmarks often overstate the performance of the real applications [26] To obtain results even closer to real applications, compact application benchmarks have also been proposed. In these, small real applications are used. These ....

F. M. McMahon, "The Livermore FORTRAN kernels: A computer test of numerical performance range," Tech. Rep. UCRL-55745, Lawrence Livermore National Laboratory, University of California, Livermore, CA, 1986.

....of artificial kernels intended to represent the computationally intensive part of certain scientific codes. The timing of this benchmark on different machines was compared and reported in terms of a unit called Whetstone instructions per second. Livermore Loops: The Livermore Fortran Kernels [7] are a set of 24 loop kernels extracted from operational codes used at Lawrence Livermore National Labs. These became widely used and publicized the concept of the Mflop unit. For scientific codes, speed measured in floating point operations per second is useful, and has less dependence on the ....

F.K.McMahon, The Livermore Fortran Kernels: a Computer Test of Numerical Performance Range, Lawrence Livermore National Lab., Technical Report UCRL-53745, 1986.

....speedup of 120 . III. USING DELAYED ADDITION IN A FLOATING POINT ACCUMULATOR Multiply and accumulation also appears frequently in floating point applications. For example, of the 24 Livermore Loops, 5 loops (loop 3, 4, 6, 9, 21) are basically long vector inner product like computation [17]. In certain applications, such as the conjugate gradient example in Section V, multiply and accumulation dominates the whole computation process. Thus it would be ideal if we could also use our delayed addition techniques to build a floatingpoint multiply and accumulator to speed up this kind of ....

McMahon, F. M. "The Livermore FORTRAN kernels: A computer test of numerical performance range," Tech. Rep. UCRL55745, Lawrence Livermore National Laboratory, Univ. of California, Livermore, December 1986.

....suites have been used to study performance. Ad hoc collections of programs have been used, more typically in early studies, as well as groups of programs explicitly designed as benchmarks. The most widely used benchmarks for processor performance are: the SPEC suite [184] the Livermore Loops [113], and for scientific machines: the Linpack Kernels. More recently the use of the PERFECT Club benchmarks [39] is also starting to appear. This work uses the Livermore Loops for primarily two reasons: 1) The Livermore Loops have been widely reported in the literature and many of the machine ....

....from scientific applications. They are intended to cover the range of common Fortran program constructs found in these types of applications. The Livermore Loops have in general shown good correlation between their performance and the performance of the scientific codes they were designed to mimic [113]. The original 14 Loops were criticized because they were found to be more heavily weighted toward vectorizable codes than a typical scientific application. Because supercomputers tend to be vector architectures and have vastly superior performance on vector codes, this would tend to over predict ....

F. H. McMahon, The Livermore Fortran Kernels: A Computer Test of the Numerical Performance Range, Lawrence Livermore National Laboratory, Technical UCRL-53745, December 1986.

....is being studied as part of the Impulse project [11] Fig. 3 lists the kernels used to generate the results presented here. daxpy, copy, and scale are from the BLAS (Basic Linear Algebra Subroutines) 18] and tridiag is a tridiagonal gaussian elimination fragment, the fifth Livermore Loop [35]. vaxpy denotes a vector axpy operation that occurs in matrixvector multiplication by diagonals: A vector a multiplied by a vector x plus a vector y. For our purposes, the actual computation in these loops is unimportant; we focus instead on the access pattern and the lengths of the streams. ....

F.H. McMahon, The Livermore Fortran Kernels: A Computer Test of the Numerical Performance Range, Technical Report UCRL-53745, Lawrence Livermore Nat'l Laboratory, Dec. 1986.

....portions, or kernels, are extracted and combined into a single benchmark program. Examples include the NAS Kernel Benchmark Program (Bailey and Barton 1985) to be described in the following paragraphs, and the Livermore Fortran Kernels, commonly referred to as the Livermore Loops (McMahon 1986). Although the kernel approach appears to be somewhat more representative than the use of synthetic programs, there are several problems with its application: Chapter 1 Introduction 5 a. Many kernel programs have been developed by research laboratories to model their vector supercomputer ....

McMahon, F. (1986). "The Livermore Fortran kernels: computer test of the numerical performance range," Technical Report UCRL-53745, Lawrence Livermore National Laboratory, Livermore, California.

....Programming Interface) 11] which supports Active Message. Active Message[8] is a fast message handling scheme with the address of the message handler contained in the message header. 3 examples have been used for experiments: LLL1(Lawrence Livermore Loop 1) LLL7(Lawrence Livermore Loop 7)[12], and Jacobi. Each of them has been programmed in C. There are two versions of each program: non overlapping and overlapping. There is a 1 dimensional array, Z, in Livermore Loop 1. The array indices of Z in the loop body are k 10 and k 11, therefore, M c = 11 and m c = 0. For a given loop ....

F. H. McMahon, "The Livermore Fortran Kernels: a Computer Test of the Numerical Performance Range," Lawrence Livermore National Laboratory, UCRL53745, UC Livermore, 1986.

....representations are important since high performance compilers spend a significant amount of compilation time scheduling operations, and thus testing for potential resource contentions. When a benchmark suite of 1327 loops from the Perfect Club [13] SPEC 89 [91] and the Livermore Fortran Kernels [65] is scheduled for the Cydra 5 machine [11] 27] approximately 50 of the total time is spent modeling the resources (i.e. answering queries such as can this operation be scheduled in this cycle ) the other 50 of the total time is spent scheduling operations (i.e. deciding the order in which ....

....machines indicate potentially 4.0 to 6.9 times faster contention queries, while requiring 22 to 67 of the memory storage used by the original machine descriptions. Dynamic measurements obtained when scheduling 1327 loops from the Perfect Club [13] SPEC 89 [91] and the Livermore Fortran Kernels [65] for the Cydra 5 machine [11] indicate that the essential work performed by the contention queries decreases by a factor of 2.76 to 3.30, depending on the functionality required by the scheduler. This decrease in essential work results, in turn, in a 1.72 to 1.83 faster execution time of the ....

[Article contains additional citation context not shown here]

F. H. McMahon. The Livermore Fortran Kernels: A computer test of the numerical performance range. Technical Report UCRL-53745, Lawrence Livermore National Laboratory, Livermore, California, 1986.

....24 loops selected from assorted scientific programs such as: Whetstone [Wic73] is a synthetic program created by measuring scientific programs written in Algol 60. This program has been converted to Fortran and has been widely used to characterize scientific program performance. Livermore Loops [McM86] are a set of kernels consisting of loops from real Fortran programs. Linpack [DMBS79] is a collection of linear algebra routines; the Gaussian elimination portion of Linpack is the segment used as a benchmark. SPEC (System Performance Evaluation Cooperative) Uni89] is a collection of ....

F.H. McMahon. The Livermore FORTRAN kernels: A computer test of the numerical performance range. Technical report, Lawrence Livermore Laboratories, December 1986.

No context found.

F. McMahon. The Livermore Fortran Kernels: A computer test of the numerical performance range. Technical Report UCRL-53745, Lawrence Livermore National Laboratory, December 1986.

No context found.

F. McMahon. The Livermore Fortran Kernels: A computer test of the numerical performance range. Technical report, Lawrence Livermore National Laboratory, December 1986.

No context found.

MCMAHON, F. The livermore fortran kernels: A computer test of the numerical performance range. Tech. Rep. UCRL-53745, Lawrence Livermore National Laboratory, December 1986.

No context found.

F. McMahon. The Livermore Fortran Kernels: A computer test of the numerical performance range. Technical report, Lawrence Livermore National Laboratory, December 1986.

No context found.

F. McMahon, "The Livermore Fortran kernels: A computer test of the numerical performance range," Lawrence Livermore National Laboratory, Livermore, CA, Tech. Rep. UCRL-53745, Dec. 1986. 130

No context found.

McMahon, F.H., "The Livermore Fortran Kernels: A Computer Test of the Numerical Performance Range", Lawrence Livermore National Laboratory, UCRL-53745, December 1986. 125

No context found.

McMahon, F.H., "The Livermore Fortran Kernels: A Computer Test of the Numerical Performance Range", Lawrence Livermore National Laboratory, UCRL-53745, December 1986.

No context found.

McMahon, F. H. The Livermore Fortran kernels: A computer test of the numerical performance range. Tech. Rep. UCRL-53745, Lawrence Livermore National Laboratory, Dec. 1986.

No context found.

F. McMahon. The LivermoreFortran Kernels: A Computer Test Of The Numerical Performance Range. Lawrence Livermore National Laboratory, Livermore, California, December 1986.

No context found.

F.H. McMahon. The Livermore FORTRAN kernels: A computer test of the numerical performance range. Technical report, Lawrence Livermore Laboratories, December 1986.

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