About Alignment in structures

[Rainstorm]

hi,

when stuff is aligned in STRUCTURES that affects offsets at which the individual elements reside in memory & the memory access is that right ?

so is there any soultion to this.. or anyway to go about doing this, like i mean if i were accessing the fields through some addressing mode (word ptr [edx+2] etc), cause there would be padded bytes in some places

t  y

[u]

MASM doesn't auto-align struct members, so there are no problems :)

You do the padding manually, like:

Code Select Expand


object struct
someByte db ?
padding1 db ? ; this
someWord1 dw ?
someWord2 dw ?
padding2 dw ? ; this
objectPtr dd ?
...


[Vortex]

Hi Rainstorm,

A classical example of alignment is the usage of binary resource templates. A structure defining a template must be DWORD aligned.

[jj2007]

MASM doesn't auto-align struct members, so there are no problems :)
You do the padding manually, like:

Here is one example that works:

Code Select Expand

MyAccels ACCEL <FCONTROL or FVIRTKEY, VK_B, 201> ; Bold
ACCEL <FCONTROL or FVIRTKEY, VK_I, 202> ; Italics
ACCEL <FCONTROL or FVIRTKEY, VK_F, 203> ; Find

However, that structure is defined as follows in windows.inc:

Code Select Expand

ACCEL STRUCT WORD
  fVirt BYTE ?
  key   WORD ?
  cmd   WORD ?
ACCEL ENDS

So MASM provides for padding and alignment... but I would not fully rely on it. Test it case by case.

[MichaelW]

MASM does not automatically align structure members as the C/C++ compilers do:

http://msdn2.microsoft.com/en-us/library/aa290049.aspx

With WORD alignment specified, the ACCEL structure is aligned as the compilers would align it, with the key and cmd members aligned on their natural boundary, and this is what the API functions expect. If the WORD alignment is not specified, the default alignment is BYTE so the second and third members end up misaligned by one byte, and this misalignment causes my test code to fail.

[ToutEnMasm]

For structures alignment use ml /Zp4, it's a defaut alignment that solve many problems.

[Rainstorm]

hi.

Thanks everyone for the feedback,  links,  & examples.
That was very helpful
-

[jj2007]

For structures alignment use ml /Zp4, it's a defaut alignment that solve many problems.

How would /Zp4 behave if you had a structure with three words?

[evlncrn8]

it would pad the struct with an additional byte making the struct 4 bytes in size i would imagine...

[MichaelW]

The /Zp4 option works correctly, but only because "the fields are aligned to the minimum of the field's size and the alignment." Because of this /Zp8 would be a better choice because it will duplicate the action of the compilers for fields of 8 bytes, as well as fields of 1, 2, and 4 bytes.

Code Select Expand


; «««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««
    include \masm32\include\masm32rt.inc

    _ACCEL STRUCT
      fVirt DB ?
      key   DW ?
      cmd   DW ?
    _ACCEL ENDS

    _TEST STRUCT
      dummy DB ?
      quad  DQ ?
    _TEST ENDS

; «««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««
    .data
      accel ACCEL <>
      align 4
      _accel _ACCEL <>
      align 8
      _test _TEST <>
    .code
; «««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««
start:
; «««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««
    print ustr$(OFFSET accel.fVirt),13,10
    print ustr$(OFFSET accel.key),13,10
    print ustr$(OFFSET accel.cmd),13,10,13,10

    print ustr$(OFFSET _accel.fVirt),13,10
    print ustr$(OFFSET _accel.key),13,10
    print ustr$(OFFSET _accel.cmd),13,10,13,10

    print ustr$(OFFSET _test.dummy),13,10
    print ustr$(OFFSET _test.quad),13,10,13,10

    inkey "Press any key to exit..."
    exit
; «««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««
end start


Result with no alignment option:

Code Select Expand


4206592
4206594
4206596

4206600
4206601
4206603

4206608
4206609


Result with /Zp8 option:

Code Select Expand


4206592
4206594
4206596

4206600
4206602
4206604

4206608
4206616


BTW, there is no /Zp8 option listed in the MASM 6.0 reference, but it obviously works with 6.14.

[jj2007]

The original ACCEL structure works with all Zp settings (the minimum is Zp1). The modified _ACCEL fails with /Zp1 but works with /Zp8. The key for understanding this is indeed in the phrase you quoted:

"the fields are aligned to the minimum of the field's size and the alignment"

i.e. the fVirt DB ? gets
no padding byte with Zp1
one padding byte with Zp1
one padding byte with Zp8

Thanks for clarifying this, Michael. I was actually afraid that the whole 5-byte or 6-byte structure would get padded to 8 byte alignments, which is not the case.

Code Select Expand


;Testkey with original ACCEL, WORD
; MyAccels ACCEL <FCONTROL or FVIRTKEY, VK_B, 200>
; _ACCEL <FCONTROL or FVIRTKEY, VK_B, 201> ; Bold
; _ACCEL <FCONTROL or FVIRTKEY, VK_I, 202> ; Italics
; _ACCEL <FCONTROL or FVIRTKEY, VK_F, 203> ; Find
; _ACCEL <FCONTROL or FVIRTKEY, VK_S, 204> ; Save
; _ACCEL <FVIRTKEY, VK_F3, 205> ; Repeat find

; ....

push esi
push ebx

mov esi, offset MyAccels
mov ebx, offset ClBuffer
invoke lstrcpy, ebx, chr$("Test:")

invoke lstrcat, ebx, chr$(crlf,"A1=")
xor eax, eax
add ax, word ptr[esi+4]
invoke lstrcat, ebx, str$(eax)
add esi, 6

invoke lstrcat, ebx, chr$(crlf,"A2=")
xor eax, eax
add ax, word ptr[esi+4]
invoke lstrcat, ebx, str$(eax)
add esi, 6

invoke lstrcat, ebx, chr$(crlf,"A3=")
xor eax, eax
add ax, word ptr[esi+4]
invoke lstrcat, ebx, str$(eax)
add esi, 6

invoke lstrcat, ebx, chr$(crlf,"A4=")
xor eax, eax
add ax, word ptr[esi+4]
invoke lstrcat, ebx, str$(eax)

invoke MessageBox, 0, ebx, addr AppName, MB_OK

pop ebx
pop esi