Performance considerations CALLNAT vs. internal Subroutine vs. Function

, ,
#1

Hello all,

of course I expected that CALLNAT is slower than an internal Sub. But I never thought that the difference is that big:

My test:

define data local
01 #loops (I4) const <1000000>
01 #a  (A) dynamic init <'foo   '>
01 #i4 (I4)
01 #timx (T)
end-define
*
define prototype #trim1
  returns (A) dynamic
  define data parameter
  1 #A     (A) dynamic by value
  end-define
end-prototype
*
#timx := *TIMX
for #i4 = 1 to #loops
  reset initial #a
  #a := *TRIM(#a)
end-for
#timx := *TIMX - #timx
write #TIMX (EM=II:SS.T) 'trim'
*
#timx := *TIMX
for #i4 = 1 to #loops
  reset initial #a
  perform trim1
end-for
#timx := *TIMX - #timx
write #TIMX (EM=II:SS.T) 'internal sub'
*
#timx := *TIMX
for #i4 = 1 to #loops
  reset initial #a
  callnat 'trim1n' #a
end-for
#timx := *TIMX - #timx
write #TIMX (EM=II:SS.T) 'callnat'
*
#timx := *TIMX
for #i4 = 1 to #loops
  reset initial #a
  #a := #trim(<#a>)
end-for
#timx := *TIMX - #timx
write #TIMX (EM=II:SS.T) 'func'
*
define subroutine trim1
#a := *TRIM(#a)
end-subroutine
end

while tim1n is:

define data parameter
1 #a (A) dynamic     
end-define           
#a := *TRIM(#a)      
end

and trim17 is:

define function #trim  
  returns (A) dynamic  
  define data parameter
  1 #a (A) dynamic     
  end-define           
  #trim := *TRIM(#a)   
end-function           
end

Result:

00:02.2 trim
00:02.6 internal sub
00:26.3 callnat
00:27.5 func

And defining a prototype or not doesn’t have any influence on runtime.

#2

Some other runtimes:

00:30.2 callnat by val
00:29.7 func by val
00:24.4 external sub
02:39.2 multi purpose func with parameter (i.e. #a := #adump(<#a,‘mdb’>) )

#3

Hi Matthias;

Actually, it is worse than you might think.

Add an additional loop to your program:

#timx := *TIMX
for #i4 = 1 to #loops
reset initial #a
end-for
#timx := *TIMX - #timx
write #TIMX (EM=II:SS.T) ‘for loop time’

This will give you the “overhead” time for doing the test. To really compare percentage differences, you should subtract the overhead time from every other time. The percentage differences will now increase.

steve

#4

You’re right of course. There is some overhead time. And maybe there’s another thing: I don’t know if reset initial of a dynamic field without changing the field content/length is faster than the one we do in the trim-loops.