An Old Benchmark

[Lubos Svoboda]

Hi,

code placed bellow is an an old benchmark (100x Gammix). I have not PB/CC or PB/Win yet. Can you somebody try this code in PB/DOS, PB/CC and PB/Win (on same machine) for comparison? The same test in VB will welcome too. Please, specify results in sec and power of used computer. Thank beforehand.

Lubos

DIM A(100)
DIM B(30)
DEFINT I-L
FOR I = 1 TO 100
A(I) = I
NEXT I
FOR I = 1 TO 30
B(I) = I
NEXT I
X = .12345
BEGINTIME = TIMER
FOR K = 1 TO 100
FOR I = 1 TO 1000
S = A(11)
FOR J = 1 TO 10
L = 11 - J
S = S * X + A(L)
NEXT J
NEXT I
FOR I = 1 TO 33
S = 0
FOR J = 1 TO 30
S = S + A(J) * B(J)
NEXT J
NEXT I
FOR I = 1 TO 667
FOR J = 1 TO 30
B(J) = A(J) + B(J)
NEXT J
NEXT I
FOR I = 1 TO 1333
S = X
FOR J = 1 TO 5
S = (S - X / S) * .5
NEXT J
NEXT I
FOR I = 1 TO 67
S = A(1)
FOR J = 2 TO 100
IF A(J) >= S THEN S = A(J)
NEXT J
NEXT I
NEXT K
DURATION = TIMER - BEGINTIME
DURAT$ = STR$(DURATION)
PRINT DURAT$

[Mel Bishop]

On a Dell Inspriron 530S, dual core, 1596mHz, 1G memory. Ran the program a couple of times on each compiler.

PB/Dos 3.5: .1101427
                .1095622

PB/CC 4.04: .093125
                 .093375

PB/win 3.04: .093625
                  .093375

Hope this answers your question.

[Michael Mayerhoffer]

These are the best times and highest

.016 - .018 for PB 8.04 and CC 4.04 both mark .016 at best
.165 - .188 for PB DOS 3.5
.375 - .567 for VB 6.0

Fixed code up to drop in PBWIN or CC

 
#COMPILE EXE
FUNCTION PBMAIN () AS LONG
    DIM A(100) AS INTEGER
    DIM B(30)  AS INTEGER
    DIM x AS LONG
    DIM i AS INTEGER
    DIM s AS INTEGER
    DIM k AS INTEGER
    DIM l AS INTEGER
    DIM j AS INTEGER
        'DEFINT I-L
        FOR I = 1 TO 100
            A(I) = I
             NEXT I
 
            FOR I = 1 TO 30
                B(I) = I
                 NEXT I
    X = .12345
 
    BEGINTIME# = TIMER
    FOR K = 1 TO 100
        FOR I = 1 TO 1000
            S = A(11)
        FOR J = 1 TO 10
            L = 11 - J
            S = S * X + A(L)
            NEXT J
                NEXT I
        FOR I = 1 TO 33
            S = 0
        FOR J = 1 TO 30
            S = S + A(J) * B(J)
               NEXT J
                NEXT I
        FOR I = 1 TO 667
        FOR J = 1 TO 30
            B(J) = A(J) + B(J)
              NEXT J
                NEXT I
        FOR I = 1 TO 1333
            S = X
 
        FOR J = 1 TO 5
            S = (S - X / S) * .5
               NEXT J
                NEXT I
 
        FOR I = 1 TO 67
             S = A(1)
 
        FOR J = 2 TO 100
            IF A(J) >= S THEN S = A(J)
                NEXT J
                    NEXT I
                        NEXT K
 
    DURATION# = TIMER - BEGINTIME#
 
    DURAT$ = STR$(DURATION#)
 
 
    'Print DURAT$
    'MsgBox DURAT$
    'WAITKEY$
 
END FUNCTION

[Michael Mattias]

DIM A(100) AS INTEGER
    DIM B(30)  AS INTEGER
    DIM x AS LONG
    DIM i AS INTEGER
    DIM s AS INTEGER
    DIM k AS INTEGER
    DIM l AS INTEGER
    DIM j AS INTEGER

Win/32 ===>

    REGISTER i AS LONG, J AS LONG 
    DIM K AS LONG
    DIM A(30)  AS LONG
    DIM B(30)  AS LONG
    DIM x AS LONG
    DIM s AS LONG

[Michael Mayerhoffer]

FWIW no difference for INTEGER vs LONG on this. I tested both ways...was rather shocked. So I left them as Integers.

[Petr Schreiber jr]

Hi,

observed improvement using REGISTER variables as Michael Matthias proposed is quite huge - 46ms vs 31ms on AMD Sempron 64bit, 3400+ ( but ticks on 1.8 GHz ).

For k = 1000 the speed advantage grows a bit again: 484ms without registers or 297ms with registers.

Tests were performed on PB/Win 8.04

Also I would suggest to use GetTickCount or even better QueryPerformanceCounter for timing benchmarks, TIMER is good, but I trust the first two mentioned more.


Bye,
Petr

[Paul Dixon]

    DIM A(100) AS INTEGER
    DIM B(30)  AS INTEGER
    DIM x AS LONG
    DIM i AS INTEGER
    DIM s AS INTEGER
    DIM k AS INTEGER
    DIM l AS INTEGER
    DIM j AS INTEGER

The original program uses SINGLEs for many of its variable, not INTEGERs.

DEFINT I-L

All variables are SINGLE unless the name begins with a letter in the range I to L

I get times of 650ms under PBDOS3.5 and 38ms in PBCC4.

Paul.

[Lubos Svoboda]

Thank to all

Thank guys. Only a little note. Variables X, S, A() and B() are real type variables in original Gammix test (not integer).
Lubos

[Michael Mattias]

Also I would suggest to use GetTickCount or even better QueryPerformanceCounter for timing benchmarks,

A. Programs like this - a bunch of 'do nothing' loops - are not 'benchmarks' of any sort.

B. Exercises like this - trying to move a program to both a new compiler and a new operating system without optimizing for either - are a waste of time at best and silly at worst.

C. The best compiler in the world is not going to change a mediocre programmer into a superstar.

[Petr Schreiber jr]

Mr. Matthias,

A: The thread is marked as "An Old Benchmark", so the word seemed appropriate to me

B: ... or also can tell me if it is worth to purchase new version of compiler, even if PB/DOS on Windows (according to info from A View on PowerBASIC)is for most tasks enough? Here we get speed gain even with unoptimized version, isn't world looking nicer, birds singing cuter and everything just fantastic when realizing such a fact?

C: :bawling: ... and I was looking forward to paintbrush/artist


Best regards,
Petr

P.S. Why I posted this reaction, although it did not bring something super new to the discussion ... because:

( orignally put on xkcd.com )

[Petr Schreiber jr]

Thank guys. Only a little note. Variables X, S, A() and B() are real type variables in original Gammix test (not integer).

I am not sure what "Real" means in PB, so I modified code to use SINGLE, DOUBLE or EXT.

Results again for PB/Win 8.04, Sempron 64bit 3400+ (1.8 GHz real )

[b]Real done as   Variable memory size  Time elapsed in [i]benchmark[/i] ;)[/b]
SINGLE          4B                    31 ms
DOUBLE          8B                    62 ms
EXT            10B                    94 ms

Lubos, one very nice feature in PB/Win - MACROs make changing variable data type a very simple process:

MACRO Real = SINGLE   
...
' -- Although PB does not know Real type, it will automatically use SINGLE now
  DIM A(100) AS LOCAL Real
  DIM B(30)  AS LOCAL Real
  LOCAL x, s AS Real

So when I want to change all "real" variables to DOUBLE, I just change word in MACRO statement to

MACRO Real = DOUBLE

and so on...


Best regards,
Petr