February 18, 2003
-----------------
Bug-fix release.

December 9, 1997
----------------
This release is based on beta release 2 of BYTE Magazine's BYTEmark
benchmark program (previously known as BYTE's Native Mode
Benchmarks). This document covers the Native Mode (a.k.a. Algorithm
Level) tests; benchmarks designed to expose the capabilities of a
system's CPU, FPU, and memory system.

Running a "make" will create the binary if all goes well. It is called
"nbench" and performs a suite of 10 tests and compares the results to
a Dell Pentium 90 with 16 MB RAM and 256 KB L2 cache running MSDOS and
compiling with the Watcom 10.0 C/C++ compiler. If you define -DLINUX
during compilation (the default) then you also get a comparison to an
AMD K6/233 with 32 MB RAM and 512 KB L2-cache running Linux 2.0.32 and
using a binary which was compiled with GNU gcc version 2.7.2.3 and GNU
libc-5.4.38.

For more verbose output specify -v as an argument.

The primary web site is: http://www.tux.org/~mayer/linux/bmark.html

The port to Linux/Unix was done by Uwe F. Mayer <mayer@tux.org>.

The index-split was done by Andrew D. Balsa, and reflects the
realization that memory management is important in CPU design. The
original tests have been left alone, however, the tests NUMERIC SORT,
FP EMULATION, IDEA, and HUFFMAN now constitute the integer-arithmetic
focused benchmark index, while the tests STRING SORT, BITFIELD, and
ASSIGNMENT make up the new memory index.

The algorithms were not changed from the source which was obtained
from the BYTE web site at http://www.byte.com/bmark/bmark.htm on
December 14, 1996.  However, the source was modified to better work
with 64-bit machines (in particular the random number generator was
modified to always work with 32 bit, no matter what kind of hardware
you run it on). Furthermore, for some of the algorithms additional
resettings of the data was added to increase the consistency across
different hardware. Some extra debugging code was added, which has no
impact on normal runs.

In case there is uneven system load due to other processes while this
benchmark suite executes, it might take longer to run than on an
unloaded system. This is because the benchmark does some statistical
analysis to make sure that the reported results are statistically
significant, and an increased variation in individual runs requires
more runs to achieve the required statistical confidence.

This is a single-threaded benchmark and is not designed to measure the
performance gain on multi-processor machines.

For details and customization read bdoc.txt.

THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

The information in this file is old and no longer valid. It seems that
the GNU C library has caught up with Motorola's libmoto, and now
performance is just as good (or better) without libmoto. I'll include
the old notice out of historical reasons only. Currently libmoto is
available at ftp://ftp.mcg.mot.com/pub/SPS/PowerPC/software/mklinux/libmoto/,
but this is subject to change and not under my control.

February 18, 2003
Uwe F. Mayer

---------------------------------------------------------------------------

If you have a Motorola CPU or equivalent:

When linked with the 'libmoto' (floating point library from Motorola)
the results you obtain are much better. (FPU index of 0.896 versus
1.910 in one example.)

The Motorola math library is currently available at:
http://www.mot.com/SPS/PowerPC/support/rsw_customer_support/mklinux/libmoto/libmoto_reg_mkdev.html

If you have a Motorola CPU and you submit a result then please let me
know whether you used libmoto or not. Please read the file README.submit.

I do not have a Motorola CPU, and I can't help you with installing the
library either.

December 3, 1997
Uwe F. Mayer

December 3, 1993
================

DEC Unix 4.0 or DEC OSF1 and gcc
--------------------------------
Compiles cleanly if you don't use -funroll-loops with gcc-2.7.2.3 or earlier

DEC UNIX 4.0 or DEC OSF1 and cc
-------------------------------
CC = cc
CFLAGS = -O3
LINKFLAGS = -s -non_shared

Compiles cleanly.

SunOS and gcc
-------------
Compiles cleanly

SunOS and cc
------------
CC = cc
CFLAGS = -O3 -s

Compiles with one warning during compilation of nbench1.c

"/usr/ucbinclude/strings.h", line 48: warning: identifier redeclared: strlen
        current : function() returning int
        previous: function() returning uint : "/usr/include/string.h", line 98

HP-UX and gcc
-------------
Compiles with one warning during compilation of sysspec.c

In file included from /usr/local/lib/gcc-lib/hppa1.1-hp-hpux9.05/2.7.2.1/include/malloc.h:9,
                 from sysspec.h:37,
                 from sysspec.c:37:
/usr/local/lib/gcc-lib/hppa1.1-hp-hpux9.05/2.7.2.1/include/sys/types.h:117: warning: empty declaration
/usr/local/lib/gcc-lib/hppa1.1-hp-hpux9.05/2.7.2.1/include/sys/types.h:118: warning: empty declaration

DEC Ultrix and cc
-----------------
CC = cc
CFLAGS = -O2
LINKFLAGS = -s

Compiles with a warning about the correct usage of cut when running sysinfo.sh
cut: Usage: cut [-s] [-d<char>] {-c<list> | -f<list>} file ...
cut: Usage: cut [-s] [-d<char>] {-c<list> | -f<list>} file ...

I plan on posting a digest of results in case people mail me any.
The URL will be linked to

http://www.tux.org/~mayer/linux/bmark.html

If you want to submit, then run the benchmark (use your own
compilation, I don't care with what flags or compiler, but I want all
numbers from a single benchmark run) and fill in the template as given
in the example below:

CPU                             : AMD 5x86P75 (486DX4/133MHz)
L2 CACHE                        : 256 KB
OS                              : Linux 2.0.32
C COMPILER                      : gcc 2.7.2.3
LIBC                            : libc-5.4.38
Pentium 90 INTEGER INDEX        : 1.051
Pentium 90 FLOATING-POINT INDEX : 0.450
AMD K6/233 MEMORY INDEX         : 0.337
AMD K6/233 INTEGER INDEX        : 0.238
AMD K6/233 FLOATING-POINT INDEX : 0.230

Any other format is fine as long as it contains the same info (write
"unknown" or "?" for data you don't know). For example, you could just
cut the summary from the output of nbench and mail it together with
cache, CPU, and OS info in case it is not already present. Please do
not email me the complete output of nbench, or any other unnecessarily
long email, as this just eats up my hard-disk space.  However, long
collections of results are of course welcome.

Send your result to mayer@tux.org

Uwe F. Mayer
February 18, 2003