EAR Benchmark Specifications
 1.  GENERAL

1.1.  Classification

    EAR  is  a  CPU  intensive  floating   point   scientific   C
    application program.

1.2  Description

    EAR simulates the propagation of sound in the  human  cochlea
    (inner  ear)  and  computes  a  picture  of  sound  called  a
    cochleagram. The program takes as input a sound  file  (in  a
    number  of  different  file  formats)  and  produces either a
    cochleagram  or  a   correlagram.    A   cochleagram   is   a
    representation that roughly corresponds to spectral energy as
    a function of time.  A correlagram is our  implementation  of
    Licklider's  duplex model of pitch perception.  The result is
    a two dimensional movie.  The theory  and  implementation  of
    the  cochlea  model  are  described in "Lyon's Cochlear Model
    (The Mathematica Notebook)" and the correlagram is  described
    in "A Duplex Theory of Pitch Perception".

1.3  Source/Author

    IBM Corp. sponsors this benchmark for SPEC.  The  author  for
    Ear is Malcolm Slaney at Apple Computer.

1.4  Version/Date

    Benchmark Version 1.1 6/24/91 (Modified for  SPEC).   056.ear
    was  derived  from  Lyon's  Cochlear  Model  August 11, 1989.
    Changes  for  SPEC  include  removing  all  machine  specific
    references  (e.g.  Cray) and adding validation of results and
    "filters" suggested by Malcolm Slaney.

1.5  Other Information

    This benchmark runs on all IBM  RISC  SYSTEMS/6000  machines,
    Sun machines, DECstation 3100, and MIPS M2000.

2.  PERFORMANCE

2.1.  Metrics

    The metric used is the elapsed (real)  time  as   output   by
    bin/time.  This  time  is  very  nearly equal to the CPU time
    (where CPU time is the sum of  "user"   and   "sys"   times),
    thereby proving that virtually no I/O is performed.

2.2.  Elapsed Time

    Run-time on a  RISC  SYSTEM/6000  Model  320H  is  about  410
    seconds   (your time will vary).  The SPEC Reference time (to
    3 sig. fig.) is 25500.

2.3.  Reports

    The benchmark writes its output to a file called "result.out"
    and also writes a cochleagram called "cochlea.pic".

2.4  Additional Performance Considerations

    Variations of the elapsed time between  runs  should  not  be
    more   than   5%  under  normal circumstances. This benchmark
    should be run on an idle  machine  or  in   single-user  mode
    since   elapsed   time  may  be  highly  sensitive  to  other
    activities on the machine.

3.  SOFTWARE

3.1  Language

    EAR in implemented in C.

3.2  Operating System

    This benchmark runs on a variety  of  UNIX-based  systems.

3.3  Portability

    The benchmark runs on both 4.2 BSD   and   System   V   based
    machines. No OS dependencies exist. EAR is easy to port.

3.4  Vectorizability/Multi-Processor Issues

    Don't Know

3.5  Miscellaneous Software

    None needed for execution of the benchmark.

3.6  Known Bugs

    None.

3.7  Additional Software Considerations

    None.

4.  HARDWARE

4.1  Memory

    The performance of this benchmark is  not  memory  dependent.
    text      data     bss
    39652   17932   11900

4.2  Disks

    No special requirements.

4.3  Communication

    None.

4.4  Special Hardware

    This application is floating point  intensive  and  therefore
    would best benefit from floating point hardware support.

4.5  Additional Hardware Considerations

    None

5.  OPERATIONAL

5.1  Disk Space

    Total disk space required is < 5 MB. Size  of  executable  is
    about 96 KB.

5.2  Installation

    Should already be installed.

5.3  Execution

    Compile at the highest level of optimization possible.

5.4  Correctness Verification

    The result file "err.out" is compared by "spiff -a1e-7".

5.5  Additional Operational Considerations

    None.

5.6  Sample Run

                        Make -f M.vendor

    See "result.ref" directory for sample output.