A musical note from QuickBASIC with XP?

[Ralph]

By adding the xx=xx+1 in the For...Next loop, you're making it work twice as hard as the others. It's already counting, where the others aren't, ya know?

Just as I figured! Twice as hard, because the loop has to loop through two lines, instead of one, since the first line is passed through only once, the first time around. And, the results bear this out, giving 6 seconds for the two lines, and 3 seconds for one.

Of course, I used n# =10,000,000, in order to get some meaningful time.

From all the above, I see that any lines that can be deleted or compressed will have a measureable time-reduction on a very large number of iterations, which is what I think the Dr's reason for measuring loop-times was.

So, it all boils down to, "reduce the number of lines of code as much as possible. in any many-times iterative loop". Reducing the time in that manner, rather than the type of loop used, seems to be the significant conclusion, so far.

[Dr_Davenstein]

Exactly. Rel and I were attemptig to make a meaningful 3D engine with QB4.5, before FreeBASIC was released. Thankfully, we didn't waste a bunch of time. Even with Assembler, QB is much too slow to produce a fast 3D engine. It's ok for models with a very low amount of vertices, but when you try to add shadow extrusion, bump mapping, etc... it's just too slow.

[Ralph]

wow!!! I tried my program in FB, and, even using n# = 100000000,
and, instead of using one 10-cycle loop, I tried two 100000000-cycle loops, and I was just barely getting some decimal places for the slowest-running WHILE/WEND loop!!! FB is faster then Superman!

[Skyler]

Not faster than an Intrepid-class ship, though.

[Ralph]

Antoni:

After I ran my original program in FB, without modifications, and thought it was super, super, super fast, I got to thinking, "Maybe I have to dimension my variables first, otherwise, they are not recognized, and, the program executes extremely fast, jumping all the loops!" So, I modified the original program by dimensioning the variables first, and multiplying the constant, n#, by 150. That seems to be the approximate speed multiplier using FB versus QB.

These are the FB answers for version 0.14:
FOR/TO 3.132, FOR/TO+ 3.202, DO/LOOP 5.823, WHILE/WEND 5.794
Seems the FOR/TO loops ran almost at the same speed, and almos twice as fast as the other loops.

Here is my FB code:

Code:

'LoopTimeFB times various loops in FreeBasic, version 0.14

CLS
dim i as long, J as long, k as long, t as single, t0 as single, t1 as single
dim  t2 as single, t3 as single, t4 as single
PRINT
A$ = "Comparison of average times in 10 loops of three loop-times in FreeBasic"
PRINT SPC(40 - LEN(A$) / 2); A$
PRINT
A$ = "Second LOOP, FOR/NEXT+, has the unnecessary line, i = i + 1"
PRINT SPC(40 - LEN(A$) / 2); A$
PRINT
PRINT " Iteration            T I M E      I N      S E C O N D S            "
PRINT "           ----------------------------------------------------------"
PRINT "    No.      FOR/NEXT       FOR/NEXT+       DO/LOOP      WHILE/WEND  "
PRINT " --------- -------------  -------------  -------------  -------------"

n# = 10000000*150

FOR J = 1 TO 10
  LOCATE 9 + J, 6: PRINT J


  'FOR/NEXT
  i = 0
  t0 = TIMER
  FOR k = 1 TO n#
  NEXT k
  t = TIMER
  t1 = t1 +t - t0
  LOCATE 9 + J, 4 + 12 - 5: PRINT t1 / J


  'FOR/NEXT+
  i = 0
  t0 = TIMER
  FOR k = 1 TO n#
    i = i + 1
  NEXT k
  t = TIMER
  t2 = t2 + t - t0
  LOCATE 9 + J, 4 + 24-2: PRINT t2 / J


  'DO/LOOP
  i = 0
  t0 = TIMER
  DO WHILE i < n#
    i = i + 1
  LOOP
  t = TIMER
  t3 = t3 + t - t0
  LOCATE 9 + J, 4 + 36 + 1: PRINT t3 / J


  'WHILE/WEND
  i = 0
  t0 = TIMER
  WHILE i < n#
    i = i + 1
  WEND
  t = TIMER
  t4 = t4 + t - t0
  LOCATE 9 + J, 4 + 48+4: PRINT t4 / J

NEXT J

for i = 1 to 2:i=0:next i