SPEC Benchmark Specifications - 094.fpppp









    Fpppp is a double precision floating point FORTRAN benchmark.  It is

    difficult to vectorize because it contains very  large  basic


    blocks.






    Fpppp is a quantum chemistry benchmark which measures perfor-


    mance  on  one  style  of  computation (two electron integral

    derivative) which occurs in the  GaussianXX  series  of  pro-


    grams.  It does very little I/O.  The input to the program is

    the number of atoms.  The amount of computation to be done is


    proportional  to  the  fourth  power  of the number of atoms.

    Fpppp has been run on machines ranging from PC's to supercom-


    puters.   A  discussion of the algorithm implemented in fpppp

    may be found in an article by H.B.Schlegel, Journal of Chemi-


    cal Physics, volume 77, page 3676 (1982).






    H.B. Schlegel, Wayne State University, Detroit, Michigan.















    Elapsed time in seconds.







    Elapsed time.  On most systems, the sum of  system+user  time

    should  be  very close to the elapsed time, as this benchmark


    does little I/O. The SPEC reference time (to 3 sig. fig.) is
  
    9200 seconds.






    Emits a short table of floating point values and  an  integer


    count  of  the number of calls to FMTGEN.  The floating point

    results should agree exactly except for a  possible  rounding


    error  in  the last place.  The number of calls to FMTGEN may

    vary by 1 or 2 %.














FORTRAN







    Many.  See next section.






    Fpppp has run on Unix machines, VMS machines, and others  in-


    cluding Macintosh and Cray XMP.






    Difficult if not impossible to vectorize.
























    The executable should be well under 1MB.  Optimized  compila-


    tion may require substantial swap space (because of the large

    basic blocks).



































    time ./fpppp < ./input.ref/natoms







    Floating point results should agree exactly except for a pos-

    sible  rounding error in the last place.  The number of calls


    to FMTGEN may vary by 1 or 2 %.