What am I doing wrong?

[julioposa2]

Hi,

I have found a strange behavior in masm32, and that happens when using offset subtraction in the parameters passed in an invoke instruction.

For example:

Code Select Expand

invoke MySub, OFFSET label1 - OFFSET label2

generates different code to:

Code Select Expand

push OFFSET label1 - OFFSET label2
call MySub

The difference is that the pushed value in the first example is not correctly calculated by masm32. Using equates does not fix the issue.

Am I doing something wrong or is this a bug?

Thank you very much in advance

[Vortex]

Hi julioposa2,

Welcome to the forum.

Running the test code below, I cannot see any problem. Would you please post here your code so we can reproduce the issue?

Code Select Expand


.386
.model flat,stdcall
option casemap:none

include     \masm32\include\windows.inc
include     \masm32\include\kernel32.inc
include     \masm32\include\msvcrt.inc

includelib  \masm32\lib\kernel32.lib
includelib  \masm32\lib\msvcrt.lib


MySub       PROTO :DWORD

.data

label1  db 'abcd',0
label2  db 'efgh',0

message db 'eax = %X',13,10,13,10,0

.data?

.code

start:

    invoke  MySub,OFFSET label1 - OFFSET label2

    push    OFFSET label1 - OFFSET label2
    call    MySub

    invoke  ExitProcess,0

MySub PROC param:DWORD

    invoke  crt_printf,ADDR message,param
    ret

MySub ENDP

END start


The ouput is :

Code Select Expand


eax = FFFFFFFB

eax = FFFFFFFB

[julioposa2]

Hi Vortex, thank you very much for your prompt reply.

I think the issue only happens when referencing labels from inside the code. I was unable to reproduce the problem with your code, but modify it in this way and tell me the results:

Code Select Expand

.386
.model flat,stdcall
option casemap:none

include     \masm32\include\windows.inc
include     \masm32\include\kernel32.inc
include     \masm32\include\msvcrt.inc

includelib  \masm32\lib\kernel32.lib
includelib  \masm32\lib\msvcrt.lib


MySub       PROTO :DWORD

.data

label1  db 'abcd',0
label2  db 'efgh',0

message db 'eax = %X',13,10,13,10,0

.data?

.code

start:
    jmp cls
cls:
    invoke  MySub,OFFSET cls - OFFSET start

    push    OFFSET cls - OFFSET start
    call    MySub

    invoke  ExitProcess,0

MySub PROC param:DWORD

    ;invoke  crt_printf,ADDR message,param
    ret

MySub ENDP

END start

The value pushed in the invoke sentence is 0Ah, while the value pushed in the manual push sentence is 2.

Best regards

[clive]

Appears to be a phase error between passes in MASM 6.15

MASM 10.00 doesn't even like your syntax, prefering

Code Select Expand


    mov eax,OFFSET cls - OFFSET start
    invoke  MySub,eax


[julioposa2]

Thank you, I thought I was getting mad  :P

What masm version do you refer to? I download from this page the masm32 package version 10 and the ml assembler that comes within it is version 6.14.8444

Regards

[clive]

I typically use MASM 6.15 for most general 16 and 32-bit code, I'd probably use something else for SSE2,3,4 code.

Code Select Expand

Microsoft (R) Macro Assembler Version 6.15.8803

Code Select Expand

Microsoft (R) Macro Assembler Version 10.00.30319.01

[julioposa2]

Thanks.
I though the v10 package from the main page contained all components up to date. Why is the 6.14 version still packaged in it?
Is the latest version publicly available anywhere?

Thanks

[clive]

I though the v10 package from the main page contained all components up to date. Why is the 6.14 version still packaged in it?
Is the latest version publicly available anywhere?

Probably licencing issues, I bought my earlier versions. Plus the new version probably breaks some of the older libraries or macros. All the MASM 6.1x versions are quite serviceable.

Microsoft has MASM 10.00 posted on their website for download, perhaps with another package. Try Google or Bing?

[Vortex]

Hi julioposa2,

I use the same version 6.14.8444

[julioposa2]

Confirmed that the same issue happens in ml version "6.15.8803".
Tested also in 8.x and 10.x and ml refuses to assemble those instructions as commented by Clive.

Vortex, do you reproduce the issue with the offsets in the code section as I suggested before?

Thanks

[japheth]

It's clearly a Masm v6 bug. It happens when a jump instruction is between the two code labels.

Masm v8+ is better, because at least it complains and doesn't silently produce wrong code. As for the PUSH, there exists a workaround for Masm v8:

Code Select Expand


     PUSHD label2 - label1   ;using PUSHD makes Masm v8 accept the expression


[Vortex]

Hi julioposa2 ,

I can reproduce the issue. Here is the result :

Code Select Expand


eax = A

eax = 2


[TASMUser]

Hi Vortex, thank you very much for your prompt reply.

I think the issue only happens when referencing labels from inside the code. I was unable to reproduce the problem with your code, but modify it in this way and tell me the results:

Code Select Expand

.386
.model flat,stdcall
option casemap:none

include     \masm32\include\windows.inc
include     \masm32\include\kernel32.inc
include     \masm32\include\msvcrt.inc

includelib  \masm32\lib\kernel32.lib
includelib  \masm32\lib\msvcrt.lib


MySub       PROTO :DWORD

.data

label1  db 'abcd',0
label2  db 'efgh',0

message db 'eax = %X',13,10,13,10,0

.data?

.code

start:
    jmp cls
cls:
    invoke  MySub,OFFSET cls - OFFSET start

    push    OFFSET cls - OFFSET start
    call    MySub

    invoke  ExitProcess,0

MySub PROC param:DWORD

    ;invoke  crt_printf,ADDR message,param
    ret

MySub ENDP

END start

The value pushed in the invoke sentence is 0Ah, while the value pushed in the manual push sentence is 2.

Best regards

You are using two different segment declarations: ".code" and ".data". MASM/TASM will take such declarations very seriously.
Try to put constants into the ".code" segment.

[japheth]

You are using two different segment declarations: ".code" and ".data". MASM/TASM will take such declarations very seriously.

True. However, nobody did claim that Masm/Tasm will ignore those declarations. Anyway, it's off-topic, because the two labels are in the very same segment.

Try to put constants into the ".code" segment.

The labels (="relocatable constants") are in the .code segment already.

[Vortex]

Poasm Version 6.00.4 creates the correct code :

Code Select Expand


eax = 2

eax = 2