HP Prime/Commands/Guides/Time Testing

De TI-Planet Wiki
Aller à la navigation Aller à la recherche
This is a backup copy of the article from the old "HPWiki" of TI-Planet (that only hosted HP Prime info)

Want to time-test HPPPL commands? Luckily, there's a simple way to do it, employing TEVAL.

The Time Testing Program

LOCAL n,m,v;

f1()
BEGIN
 // THE NIL FUNCTION
END;

f2()
BEGIN
 // YOUR FUNCTION HERE
END;

loop1()
BEGIN
 LOCAL c;
 FOR c FROM 1 TO n DO
  f1()
 END;
END;

loop2()
BEGIN
 LOCAL c;
 FOR c FROM 1 TO n DO
  f2()
 END;
END;

EXPORT TIMETEST_EX(nn,mm,vv)
BEGIN
 n := nn;
 v := vv;
 LOCAL a := {};
 LOCAL c;
 FOR m FROM 1 TO mm DO
  LOCAL ms1 := TEVAL(loop1());
  LOCAL ms2 := TEVAL(loop2());
  LOCAL tstn := ms2/ms1;
  a[0] := tstn;
 END;
 CAS.median(a);
END;

EXPORT TIMETEST(vv)
BEGIN
 RETURN TIMETEST_EX(10000,100,vv);
END;

EXPORT TIMETEST_SIMPLE()
BEGIN
 RETURN TIMETEST(0);
END;

Using The Program

Inside the body of function f2, write the command you wish to time-test. 'v' is an argument that you can feed into your program (see below).

TIMETEST_EX is the most flexible of the functions. It takes 3 arguments:

  • The number of looped executions, 'n'
  • The number of time-tests to average, 'm'
  • The arbitrary value to pass in to f2, 'v'

TIMETEST is like TIMETEST_EX, but provides default values for 'n' (10000) and 'm' (100). It's good for smaller programs; you will likely want to adjust the values via TIMETEST_EX for larger programs.

TIMETEST_SIMPLE is like TIMETEST, but used when programs don't need a 'v' specified for it.

Results

It will run your code n*m times, finding the average running time versus the nil case. The result is produces is in TSTN (Times Slower Than Nil). This value is how many times slower it is than running an empty function. This value is portable between real and virtual calculators. In addition, if function A has a TSTN of X, and function B has a TSTN of Y, then A is (X-Y)/Y times faster/slower than B.

Be sure to run TIMETEST more than once. If the results differ a lot between runs, increase n or m and try again.

Limitations

TSTN, while supposed to be a consistent measurement between systems, may not be entirely equivalent between virtual and real calculators, as the architecture (and therefore the compiled code) differs between the ARM and x86/x64 computers. More research has to be done into this. For now, if you're looking for TSTN, use the virtual calculator; it completes tests faster.