Test eax,eax

[Farabi]

What is test eax,eax mean? And what is it used for?

[Vortex]

Hi Onan,

test eax,eax is used to check if eax is equal to 0. If this is the case then the ZERO flag is set to TRUE.

An example :

Code Select Expand

invoke  VirtualAlloc,0,eax,MEM_COMMIT,PAGE_READWRITE            ; allocate memory
    test    eax,eax
    jnz     @f

If VirtualAlloc fails, it returns 0. If the value of eax is not zero then the ZERO flag is set to FALSE. This means that the code will jump to the nearest forward label.

From \masm32\help\opcodes.hlp :

TEST - Test For Bit Pattern
        Usage:  TEST    dest,src
        Modifies flags: CF OF PF SF ZF (AF undefined)
        Performs a logical AND of the two operands updating the flags
        register without saving the result.

[hoverlees]

TEST is Only a little different from AND.

[PBrennick]

Both of you gentlemen make correct points but the most important point is that, while AND is a destructive test; Test is not. This means, if you use AND, the destination is modified and the flags set accordingly. If you use TEST, the flags are modified accordingly but the destination variable (and the source) are not modified. TEST does an AND of the 2 variables but does not change them, in other words.

-- Paul

[jj2007]

while AND is a destructive test; Test is not

Another non-destructive test is or eax, eax

[Mark Jones]

Timings may differ on the various platforms but OR EAX,EAX and TEST EAX,EAX are generally much faster than CMP EAX,0 or similar.

Likewise, if it is necessary to perform many tests, it may be even faster to use a look-up table. There should be many posts here describing how to use a lookup table (see the XLAT instruction.)

[Farabi]

Thanks,  I understad it now.

[raymond]

while AND is a destructive test; Test is not

Another non-destructive test is or eax, eax

"or eax,eax" is as much a destructive test as the "and eax,eax".
They both entail writing the result to the register (even though there is no change in these cases) while test does not write to any general purpose register.

[donkey]

Also remember that TEST does not only set the zero flag, it will set any of the flags for example if you need to know if the number is negative or if you use it with a constant you can check the value of a bit in eax, eg...

test EAX, 1
jnz @F ; << jump if bit 0 is set

this would be equivalent to the BT opcode, except it can check multiple bits for example TEST EAX, 6 will result in a zero flag if either bit 1 or bit 2 is not set. It is a very useful opcode and a relatively fast one.

Donkey

[hutch--]

TEST is a very useful mnemonic, test any bit and not change the value in the register.

Code Select Expand


; ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤
    include \masm32\include\masm32rt.inc
; ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤

comment * -----------------------------------------------------
                        Build this  template with
                       "CONSOLE ASSEMBLE AND LINK"
        ----------------------------------------------------- *

    .code

start:
   
; ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤

    call main
    inkey
    exit

; ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤

main proc


    mov edx, 00000000000000010000000000000000b      ; set bit 17

    test edx, 00000000000000010000000000000000b     ; test bit 17
    jz notset

    print "Bit 17 is set",13,10
    jmp quit

  notset:
    print "Bit 17 is NOT set",13,10

  quit:

    ret

main endp

; ¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤¤

end start


[jj2007]

"or eax,eax" is as much a destructive test as the "and eax,eax".
They both entail writing the result to the register (even though there is no change in these cases) while test does not write to any general purpose register.

What is your definition of "destructive"?

[Tedd]

"test" does a virtual 'and' on the destination operand with the source operand, but does not write the result back.
"and" does an 'and' on the destination operand with the source operand, and writes the result back to the destination operand.
"or" does an 'or' on the destination operand with the source operand, and writes the result back to the destination operand.

(a related example is 'cmp' and 'sub' -- cmp does a virtual sub, but doesn't write the result back to the destination.)


If you "and" or "or" a value with itself, the result will be the original value (with only the flags being modified), but these instructions are 'destructive' in that they write the result back to the destination operand. They may write the exact same value back, but they still essentially destroy the previous value (it's modified, even if not changed.)



Farabi: "test eax,eax,", "and eax,eax", "or eax,eax", "cmp eax,0" are all just ways of checking if eax == 0. They don't change the value in eax, but if the result of the operation is 0, then the zero-flag is set.
If you 'and' a number with itself, you will get the same number as the result, and the zero-flag will be set if the result of an operation (i.e. that number) is zero. And the same for if you 'or' a number with itself.
"cmp" also works the same -- if you subtract 0 from a number, the result will be the same, and the zero-flag will be set if that result is zero (which must mean the original number was zero: 0 - 0 = 0).

[jj2007]

Ok, what you write is technically and philosophically correct, but let me put it more bluntly: Does it matter? Will the CPU police put me in jail if I destroy the valuable contents of eax with or eax, eax? Will the OS police hit me over the head if I insert a mov esi, esi in my callback procedure although I forgot uses esi?

[raymond]

Will the OS police hit me over the head if I insert a mov esi, esi in my callback procedure although I forgot uses esi?

This has nothing to do with the fact that "or" is destructive.

(BTW, I very often use the esi, edi, ebx (and even the ebp if I need it) registers without saving them IF they don't need to be saved. :clap:)

[Tedd]

Ok, what you write is technically and philosophically correct, but let me put it more bluntly: Does it matter?

Obviously, no - not really.
However, one could argue that's it's more efficient, since writing back a value requires an extra micro-op that could really be skipped otherwise.

Will the CPU police put me in jail if I destroy the valuable contents of eax with or eax, eax? Will the OS police hit me over the head if I insert a mov esi, esi in my callback procedure although I forgot uses esi?

We can always hope, my friend :bdg