Testing an odd number

[dedndave]

actually lingo - i was making the remark and rolling my eyes because of all your nonsense - lol

[dedndave]

Frank - if the contents of the register are "disposable", you can use SHR

Code Select Expand

        shr     eax,1
        jc      value_was_odd
the original value may be recovered, if desired with RCL EAX,1
as long as the carry flag has not been altered

[frktons]

Frank - if the contents of the register are "disposable", you can use SHR

Code Select Expand

        shr     eax,1
        jc      value_was_odd
the original value may be recovered, if desired with RCL EAX,1
as long as the carry flag has not been altered

Thanks Dave. Good to know, it's one of the solutions I tested.
What about avoiding the eax MOV and testing the variable directly?

Code Select Expand


;-------------------------------------------------------------------------------------------------------------------------
; Test if a number is ODD or EVEN
; Date: 27 august 2010
;-------------------------------------------------------------------------------------------------------------------------


INCLUDE \masm32\include\masm32rt.inc

;------------------------------------------------------------------------------------------------------------------------------

.DATA

number    dd    12


;------------------------------------------------------------------------------------------------------------------------------


.CODE


start:

    bt number, 0

    jc  isodd

    print "The number is EVEN",13,10

    inkey

    jmp end_test

isodd:

    print "The number is ODD",13,10

    inkey


end_test:


    ret

end start



[frktons]

Well testing the number directly has some small advantages:

Code Select Expand


Intel(R) Core(TM)2 CPU          6600  @ 2.40GHz (SSE4)
2072    cycles for 1000*test eax, 1, result=500
2028    cycles for 1000*bt eax, 0, result=500

2016    cycles for 1000*bt number, 0, result=1000

2041    cycles for 1000*test eax, 1, result=500
2036    cycles for 1000*bt eax, 0, result=500

2025    cycles for 1000*bt number, 0, result=1000


--- ok ---


I tested with number = 13.

[lingo]

Hi,Frank
"What about avoiding the eax MOV and testing the variable directly?"
Will be faster coz you will skip mov eax, number.

dedndave,
actually dedndave - have you some news  about your wife coz it is not healthy for a man to be alone? :snooty:

[frktons]

Hi,Frank
"What about avoiding the eax MOV and testing the variable directly?"
Will be faster coz you will skip mov eax, number.

Well, there is infact a quite small advantage as I realized with my test.
Tell me Lingo, do you really think these tests are useless, or are you
just joking?

[lingo]

Will be better for you to test it. Just make  two algos equal and test them. The times must be the same but I think they will be different... :U

[frktons]

Will be better for you to test it. Just make  two algos equal and test them. The times must be the same but I think they will be different... :U

On modern OSes, with dozens of tasks around all the time, is almost impossible
to have two equal results. I think we can only have a broad idea of the speed, and
this should be the target of the tests. Thanks anyway for your comments.  :U

[jj2007]

the original value may be recovered, if desired with RCL EAX,1
as long as the carry flag has not been altered

Cute, Dave. Didn't know that, probably since OpCodes.chm tells a different story:

Rotates the bits in the destination to the left "count" times with all data pushed out the left side re-entering on the right. The Carry Flag holds the last bit rotated out.

Nothing on carry flag rotating in etc...

But I checked it with Olly, and you are right, as usual :U

[dedndave]

i am a little surprised, Jochen, that you don't have a solid handle on SH/RO/RC instructions
that's some of the first asm stuff i learned   :P
in the beginning, i didn't have many tools - had to figure out how things worked manually
when i first got a PC, i thought DEBUG was the bees knees - lol

on an 8008, we considered CF to be the "9th bit"   :lol

[jj2007]

i am a little surprised, Jochen, that you don't have a solid handle on SH/RO/RC instructions
that's some of the first asm stuff i learned   :P

My solid handles are \masm32\help and Google, and I thought \masm32\help\opcodes.chm was enough. Usually it is correct.

[mineiro]

Hello Sr frktons;
I'm getting float results here; while at every firsts test "test" wins, but after some 10 tests the result invert and "bt" wins.
If I put some huge program to be load, while doing the test program, this results preserve here. At night I can do more tests to you.
regards.

[dedndave]

at my age, my solid handles seem to be around my waist - lol
all my life, i have worn slacks and jeans with 34 to 36 waist
had to get some new stuff - 38" now   :lol

[jj2007]

hiyas Jochen - good morning (here) - good afternoon (there)

i can't imagine that much difference in speed  :P
must be something wrong with the test   ::)   :lol

Yes, sure! On a P4, test al is three millicycles slower:

Code Select Expand

Intel(R) Pentium(R) 4 CPU 3.40GHz (SSE3)
4034    1000*test eax, 1, result=500
4037    1000*test al, 1, result=500
10676   1000*bt eax, 0, result=500

By the way: Lingo is so quiet. I think he is cooking up an SSE5 version that runs in -50 cycles - on his CPU only :green2

P.S.:
> had to get some new stuff - 38" now   :lol
My condolencies - I know the problem :thumbu

P.P.S.: For those who draw their skills from OpCode.chm, here a "second opinion" (MasmReference):

RCL/RCR/ROL/ROR Rotate
Rotates the bits in the destination operand the number of times specified in the
source operand. RCL and ROL rotate the bits left; RCR and ROR rotate right.
ROL and ROR rotate the number of bits in the operand. For each rotation, the
leftmost or rightmost bit is copied to the carry flag as well as rotated. RCL and
RCR rotate through the carry flag. The carry flag becomes an extension of the
operand so that a 9-bit rotation is done for 8-bit operands, or a 17-bit rotation
for 16-bit operands.
On the 8088 and 8086, the source operand can be either CL or 1. On the
80186–80486, the source operand can be CL or an 8-bit constant. On the
80186–80486, rotate counts larger than 31 are masked off, but on the 8088 and
8086, larger rotate counts are performed despite the inefficiency involved. The
overflow flag is modified only by single-bit variations of the instruction; for
multiple-bit variations, the overflow flag is undefined.

[bomz]

masm32 applications limits by windows procedure, and serious code optimization needs only for huge computation.



whether there Patch to remove this header ?