NAME OR DESIGNATION OF PROGRAM, COMPUTER, DESCRIPTION OF PROGRAM OR FUNCTION, METHOD OF SOLUTION, RESTRICTIONS ON THE COMPLEXITY OF THE PROBLEM, TYPICAL RUNNING TIME, UNUSUAL FEATURES OF THE PROGRAM, RELATED AND AUXILIARY PROGRAMS, STATUS, REFERENCES, MACHINE REQUIREMENTS, LANGUAGE, OPERATING SYSTEM UNDER WHICH PROGRAM IS EXECUTED, OTHER PROGRAMMING OR OPERATING INFORMATION OR RESTRICTIONS, NAME AND ESTABLISHMENT OF AUTHORS, MATERIAL, CATEGORIES

[ top ]

[ top ]

To submit a request, click below on the link of the version you wish to order.
Only liaison officers are authorised to submit online requests. Rules for requesters are
available here.

Program name | Package id | Status | Status date |
---|---|---|---|

LFK | NESC9426/01 | Tested | 02-DEC-1991 |

Machines used:

Package ID | Orig. computer | Test computer |
---|---|---|

NESC9426/01 | Many Computers | DEC VAX 8810 |

[ top ]

3. DESCRIPTION OF PROGRAM OR FUNCTION

LFK, the Livermore FORTRAN Kernels, is a computer performance test that measures a realistic floating-point performance range for FORTRAN applications. Informally known as the Livermore Loops test, the LFK test may be used as a computer performance test, as a test of compiler accuracy (via checksums) and efficiency, or as a hardware endurance test. The LFK test, which focuses on FORTRAN as used in computational physics, measures the joint performance of the computer CPU, the compiler, and the computational structures in units of Megaflops/sec or Mflops. A C language version of subroutine KERNEL is also included which executes 24 samples of C numerical computation. The 24 kernels are a hydrodynamics code fragment, a fragment from an incomplete Cholesky conjugate gradient code, the standard inner product function of linear algebra, a fragment from a banded linear equations routine, a segment of a tridiagonal elimination routine, an example of a general linear recurrence equation, an equation of state fragment, part of an alternating direction implicit integration code, an integrate predictor code, a difference predictor code, a first sum, a first difference, a fragment from a two-dimensional particle-in-cell code, a part of a one-dimensional particle-in-cell code, an example of how casually FORTRAN can be written, a Monte Carlo search loop, an example of an implicit conditional computation, a fragment of a two-dimensional explicit hydrodynamics code, a general linear recurrence equation, part of a discrete ordinates transport program, a simple matrix calculation, a segment of a Planck distribution procedure, a two-dimensional implicit hydrodynamics fragment, and determination of the location of the first minimum in an array.

LFK, the Livermore FORTRAN Kernels, is a computer performance test that measures a realistic floating-point performance range for FORTRAN applications. Informally known as the Livermore Loops test, the LFK test may be used as a computer performance test, as a test of compiler accuracy (via checksums) and efficiency, or as a hardware endurance test. The LFK test, which focuses on FORTRAN as used in computational physics, measures the joint performance of the computer CPU, the compiler, and the computational structures in units of Megaflops/sec or Mflops. A C language version of subroutine KERNEL is also included which executes 24 samples of C numerical computation. The 24 kernels are a hydrodynamics code fragment, a fragment from an incomplete Cholesky conjugate gradient code, the standard inner product function of linear algebra, a fragment from a banded linear equations routine, a segment of a tridiagonal elimination routine, an example of a general linear recurrence equation, an equation of state fragment, part of an alternating direction implicit integration code, an integrate predictor code, a difference predictor code, a first sum, a first difference, a fragment from a two-dimensional particle-in-cell code, a part of a one-dimensional particle-in-cell code, an example of how casually FORTRAN can be written, a Monte Carlo search loop, an example of an implicit conditional computation, a fragment of a two-dimensional explicit hydrodynamics code, a general linear recurrence equation, part of a discrete ordinates transport program, a simple matrix calculation, a segment of a Planck distribution procedure, a two-dimensional implicit hydrodynamics fragment, and determination of the location of the first minimum in an array.

[ top ]

4. METHOD OF SOLUTION

CPU performance rates depend strongly on the maturity of FORTRAN compiler machine code optimization. The LFK test-bed executes the set of 24 kernels three times, resetting the DO-loop controls so that short, medium, and long vector performance is sampled and can be compared. Following these three executions, the 72 timings are combined for statistical analysis and printed. The entire LFK test is executed seven times to measure experimental timing errors. An analysis of these timing errors for each kernel is provided to confirm the accuracy of the test. The LFK test also computes a sensitivity analysis of the weighted harmonic mean rate by assigning 49 sets of weights to the kernels. This analysis may be used for risk analysis to understand the variation in net performance that different workloads would cause. The LFK test report concludes with an analysis of the sensitivity of the net FORTRAN rate to optimization using the SISD/SIMD model, a two-component form of the weighted harmonic mean (harmonic Mflops) model. This analysis may be used to gauge the performance of applications from a knowledge of their vectorizability.

CPU performance rates depend strongly on the maturity of FORTRAN compiler machine code optimization. The LFK test-bed executes the set of 24 kernels three times, resetting the DO-loop controls so that short, medium, and long vector performance is sampled and can be compared. Following these three executions, the 72 timings are combined for statistical analysis and printed. The entire LFK test is executed seven times to measure experimental timing errors. An analysis of these timing errors for each kernel is provided to confirm the accuracy of the test. The LFK test also computes a sensitivity analysis of the weighted harmonic mean rate by assigning 49 sets of weights to the kernels. This analysis may be used for risk analysis to understand the variation in net performance that different workloads would cause. The LFK test report concludes with an analysis of the sensitivity of the net FORTRAN rate to optimization using the SISD/SIMD model, a two-component form of the weighted harmonic mean (harmonic Mflops) model. This analysis may be used to gauge the performance of applications from a knowledge of their vectorizability.

[ top ]

[ top ]

6. TYPICAL RUNNING TIME

LFK may use substantially more CPU time on systems with poor CPU-clock resolution because very accurate, convergent methods were developed to measure the overhead time used for subroutines SECOND and TEST in subroutines SECOVT and TICK, respectively.

LFK may use substantially more CPU time on systems with poor CPU-clock resolution because very accurate, convergent methods were developed to measure the overhead time used for subroutines SECOND and TEST in subroutines SECOVT and TICK, respectively.

NESC9426/01

The FORTRAN program was run at NEADB on a VAX-8810 computer. In single precision, CPU time was 384 seconds; in double precision, CPU time was 374 seconds.[ top ]

[ top ]

[ top ]

10. REFERENCES

- LFK, NESC No. 9426, LFK Tape Description

NESC Note 90-117 (July 31, 1990).

- John T. Feo

An Analysis of the Computational and Parallel Complexity of the

Livermore Loops

Parallel Computing, Vol. 7, No.2, pp.163-185 (1988).

- LFK, NESC No. 9426, LFK Tape Description

NESC Note 90-117 (July 31, 1990).

- John T. Feo

An Analysis of the Computational and Parallel Complexity of the

Livermore Loops

Parallel Computing, Vol. 7, No.2, pp.163-185 (1988).

NESC9426/01, included references:

- Franck H. McMahon:The Livermore Fortran Kernels: A Computer Test of the Numerical

Performance Range

UCRL-53745 (December 1986).

[ top ]

[ top ]

[ top ]

[ top ]

NESC9426/01

File name | File description | Records |
---|---|---|

NESC9426_01.001 | Information file | 80 |

NESC9426_01.002 | LFK source program (FORTRAN) | 7338 |

NESC9426_01.003 | LFK source program (C language) | 1424 |

Keywords: FORTRAN, computers, parallel processing, performance testing.