Loop optimization in QB64

Question

Have a loop in QB64 concerning loop optimization:

DIM N AS DOUBLE, X(100000000) AS DOUBLE
T! = TIMER
FOR N = 1 to 100000000
    IF X(N) THEN
        PRINT X(N)
        EXIT FOR
    END IF
NEXT
PRINT TIMER - T!

is it any faster than:

DIM N AS DOUBLE, X(100000000) AS DOUBLE
T! = TIMER
FOR N = 1 to 100000000
    IF X(N) <> 0  THEN
        PRINT X(N)
        EXIT FOR
    END IF
NEXT
PRINT TIMER - T!

EDITED: 09-18-2018 to include variable types

Answer 1

I written this code to evaluate your test:

REM Delete REM to enable console runs
REM $CONSOLE:ONLY
REM _DEST _CONSOLE

DIM SHARED N AS DOUBLE, X(100000000) AS DOUBLE

S# = 0: ZC% = 0
T% = 10
IF COMMAND$ <> "" THEN
    T% = VAL(COMMAND$)
END IF

IF T% > 999 THEN T% = 999

FOR I% = 1 TO T%
    A# = TRYA
    B# = TRYB

    D# = A# - B#
    PRINT USING "Case A ... : #.########"; A#
    PRINT USING "Case B ... : #.########"; B#
    PRINT USING "Diff ..... : #.########"; D#;

    A$ = ""
    IF ABS(D#) < 0.00000001 THEN
        ZC% = ZC% + 1
        A$ = "*"
    END IF

    S# = S# + A# - B#

    PRINT A$
    PRINT

    REM INKEY$ doesn't work in console mode!
    A$ = INKEY$
    IF A$ = CHR$(27) THEN
        I% = I% + 1: EXIT FOR
    END IF
NEXT

PRINT USING "Avrg A - B : #.########"; S# / (I% - 1)
PRINT USING "0 diff:### on ### tryes"; ZC%, (I% - 1)
PRINT

PRINT "Hit a key to exit!"
REM INPUT$ doesn't work in console mode!
A$ = INPUT$(1)
SYSTEM


FUNCTION TRYA#
    T# = TIMER
    FOR N = 1 TO 100000000
        IF X(N) THEN
            PRINT X(N)
            EXIT FOR
        END IF
    NEXT
    A# = TIMER - T#

    TRYA = A#
END FUNCTION

FUNCTION TRYB#
    T# = TIMER
    FOR N = 1 TO 100000000
        IF X(N) <> 0 THEN
            PRINT X(N)
            EXIT FOR
        END IF
    NEXT
    A# = TIMER - T#

    TRYB = A#
END FUNCTION

The two different routines are inserted into two functions: TRYA and TRYB.

I launched this SW with a loop that runs 999 times the functions and the result is:

Avrg. A - B: 0.00204501
0 diff:359 on 999 tryes

then I launched with a 10 times loop and the result is:

Avrg. A - B: -.01640625
0 diff:  1 on  10 tryes

then I launched with a 15 times loop and the result is:

Avrg. A - B: 0.00026042
0 diff:  5 on  15 tryes

Cause we launch the SW in a multi-thread ambient I don't believe this is a very good test, but there's some results:

1. In two cases the results of no difference (0 diff) is a third of all loops.
2. In two cases it seems the function TRYA is slower.
3. In one case it seems the function TRYB is slower.

Looking at these results, I think, we may consider the two functions equivalent!

You obtain more than 10 loops running the code from command line (or modifying the command$ parameter into the QB64 menu) as:

# ./test n

Where n is the number of loops you desire.

The SW was compiled using gcc with -O3 optimizations option. (To do this you have to modify the file [/opt/]qb64/internal/c/makeline_lnx.txt)