'\"
'\" Copyright (c) 2005 Sergey Brester aka sebres.
'\"
'\" See the file "license.terms" for information on usage and redistribution
'\" of this file, and for a DISCLAIMER OF ALL WARRANTIES.
'\"
.TH timerate n "" Tcl "Tcl Built-In Commands"
.so man.macros
.BS
'\" Note:  do not modify the .SH NAME line immediately below!
.SH NAME
timerate \- Calibrated performance measurements of script execution time
.SH SYNOPSIS
\fBtimerate \fIscript\fR ?\fItime\fR? ?\fImax-count\fR?
.sp
\fBtimerate \fR?\fB\-direct\fR? ?\fB\-overhead\fI double\fR? \fIscript\fR ?\fItime\fR? ?\fImax-count\fR?
.sp
\fBtimerate \fR?\fB\-calibrate\fR? ?\fB\-direct\fR? \fIscript\fR ?\fItime\fR? ?\fImax-count\fR?
.BE
.SH DESCRIPTION
.PP
The \fBtimerate\fR command does calibrated performance measurement of a Tcl
command or script, \fIscript\fR. The \fIscript\fR should be written so that it
can be executed multiple times during the performance measurement process.
Time is measured in elapsed time using the finest timer resolution as possible,
not CPU time; if \fIscript\fR interacts with the OS, the cost of that
interaction is included.
This command may be used to provide information as to how well a script or
Tcl command is performing, and can help determine bottlenecks and fine-tune
application performance.
.PP
The first and second form will evaluate \fIscript\fR until the interval
\fItime\fR given in milliseconds elapses, or for 1000 milliseconds (1 second)
if \fItime\fR is not specified.
.sp
The parameter \fImax-count\fR could additionally impose a further restriction
by the maximal number of iterations to evaluate the script.
If \fImax-count\fR is specified, the evalution will stop either this count of
iterations is reached or the time is exceeded.
.sp
It will then return a canonical tcl-list of the form:
.PP
.CS
\fB0.095977 \(mcs/# 52095836 # 10419167 #/sec 5000.000 net-ms\fR
.CE
.PP
which indicates:
.IP \(bu 3
the average amount of time required per iteration, in microseconds ([\fBlindex\fR $result 0])
.IP \(bu 3
the count how many times it was executed ([\fBlindex\fR $result 2])
.IP \(bu 3
the estimated rate per second ([\fBlindex\fR $result 4])
.IP \(bu 3
the estimated real execution time without measurement overhead ([\fBlindex\fR $result 6])
.PP
The following options may be supplied to the \fBtimerate\fR command:
.TP
\fB\-calibrate\fR
.
To measure very fast scripts as exactly as possible, a calibration process
may be required.
The \fB\-calibrate\fR option is used to calibrate \fBtimerate\fR itself,
calculating the estimated overhead of the given script as the default overhead
for future invocations of the \fBtimerate\fR command. If the \fItime\fR
parameter is not specified, the calibrate procedure runs for up to 10 seconds.
.RS
.PP
Note that calibration is not thread safe in the current implementation.
.RE
.TP
\fB\-overhead \fIdouble\fR
.
The \fB\-overhead\fR parameter supplies an estimate (in microseconds) of the
measurement overhead of each iteration of the tested script. The passed value
overrides, for the current invocation of \fBtimerate\fR, the overhead
estimated by a previous calibration. Overrides may themselves be measured
using \fBtimerate\fR as illustrated by a later example.
.TP
\fB\-direct\fR
.
The \fB-direct\fR option causes direct execution of the supplied script,
without compilation, in a manner similar to the \fBtime\fR command. It can be
used to measure the cost of \fBTcl_EvalObjEx\fR, of the invocation of canonical
lists, and of the uncompiled versions of bytecoded commands.
.PP
As opposed to the \fBtime\fR commmand, which runs the tested script for a fixed
number of iterations, the timerate command runs it for a fixed time.
Additionally, the compiled variant of the script will be used during the entire
measurement, as if the script were part of a compiled procedure, if the \fB\-direct\fR
option is not specified. The fixed time period and possibility of compilation allow
for more precise results and prevent very long execution times by slow scripts, making
it practical for measuring scripts with highly uncertain execution times.
.SH EXAMPLES
Estimate how fast it takes for a simple Tcl \fBfor\fR loop (including
operations on variable \fIi\fR) to count to ten:
.PP
.CS
\fI# calibrate\fR
\fBtimerate\fR -calibrate {}

\fI# measure\fR
\fBtimerate\fR { for {set i 0} {$i<10} {incr i} {} } 5000
.CE
.PP
Estimate how fast it takes for a simple Tcl \fBfor\fR loop, ignoring the
overhead of the management of the variable that controls the loop:
.PP
.CS
\fI# calibrate for overhead of variable operations\fR
set i 0; \fBtimerate\fR -calibrate {expr {$i<10}; incr i} 1000

\fI# measure\fR
\fBtimerate\fR {
    for {set i 0} {$i<10} {incr i} {}
} 5000
.CE
.PP
Estimate the speed of calculating the hour of the day using \fBclock format\fR only,
ignoring overhead of the portion of the script that prepares the time for it to
calculate:
.PP
.CS
\fI# calibrate\fR
\fBtimerate\fR -calibrate {}

\fI# estimate overhead\fR
set tm 0
set ovh [lindex [\fBtimerate -overhead 0\fR {
    incr tm [expr {24*60*60}]
}] 0]

\fI# measure using estimated overhead\fR
set tm 0
\fBtimerate\fR -overhead $ovh {
    clock format $tm -format %H
    incr tm [expr {24*60*60}]; # overhead for this is ignored
} 5000
.CE
.PP
In this last example, note that the overhead itself is measured using
\fBtimerate\fR invoked with \fB-overhead 0\fR. This is necessary
because explicit overheads are assumed to be absolute values,
and not an increment over the default calibrated overhead. It is
therefore important that the calibrated overhead is excluded in the
measurement of the overhead value itself. This is accomplished by
passing \fB-overhead 0\fR when measuring the overhead.
.SH "SEE ALSO"
time(n)
.SH KEYWORDS
performance measurement, script, time
.\" Local Variables:
.\" mode: nroff
.\" End: