graphene/LibOS/shim/test/apps/lmbench/lmbench-2.5/doc/mhz.8

.\" $Id$
.TH MHZ 8 "$Date$" "(c)1994 Larry McVoy" "LMBENCH"
.SH NAME
mhz \- calulate processor clock rate
.SH SYNOPSIS
.B mhz
.I [-c]
.SH DESCRIPTION
.B mhz
calculates the processor clock rate in megahertz.
Since it is written in portable ANSI/C, it cannot rely on 
the operating system to report the clock speed.
Nor can it measure the time of a single loop to determine 
the clock rate since this would require that it assume
the number of clock ticks per instruction, which is not
constant across architectures.
.P
.B mhz
measures the time to execute a variety of loops, which are
carefully constructed to minimize the compiler and processor's
ability to overlap instruction execution so that time total
time per loop is an integral multiple of clock ticks.
.B mhz
then finds the greatest common divisor of the various
time measurements to identify the clock rate.
.P
Of course, things aren't really that simple.  For one thing,
the measured times are real numbers, not integers, so we must
convert the real numbers to integers in order for the
.I "greatest common divisor"
to make sense.
In reality, 
.B mhz
makes a series of assumptions that the smallest measured
time is {1, 2, ...} clock ticks.  
.B mhz
then computes the number of clock ticks for each measured
value, and the error between the predicted value and the
measured value.
.P
.B mhz
also has a variety of mechanisms for reducing the impact
of measurement error on the reported results, since the
algorithm is extremely sensitive to measurement error.
.SH OUTPUT
Output format is either just the clock rate as a float (-c) or more verbose
.sp
.ft CB
39.80 Mhz, 25 nanosec clock
.ft
.SH BUGS
.SH ACKNOWLEDGEMENT
Funding for the development of
this tool was provided by Sun Microsystems Computer Corporation.
.SH "SEE ALSO"
lmbench(8).