Hi!

Does anyone know if it is possible to control jump translation logic in FASM? Is there some directive to define if compiler should use absolute or relative addressing?

The problem is that I need this piece of code:
...
jmp SomeLabel
SomeLabel: stc
...

to be treated like having an absolute addressing not relative. Something like this:
...
0B3A: jmp near 0B3D
0B3D: stc
0B3E: ...

Right now all I get is:
...
0B3A: jmp $+2 ;(or "jmp $+3" if I use "jmp word SomeLabel")
0B3D: stc
0B3E: ...

Thank you.

There are no direct near jumps using absolute addressing in the Intel Architecture. Or did I misunderstood something?

You're right, I need to refresh my knowledge of IA.

Thank you for giving such a quick answer
You can remove a topic if you like to.

You can always use a CALL instead of a jmp

call SomeLabel

SomeLabel:
add esp,4 ;; clean the stack from the call
...

CALL uses relative addressing the same as JMP.

If I'm not mistaken a near Call has two encodings

Call RelOffset ; IP + Offset is encoded E8 ?
Call Reg/addr ; IP = Reg/Addr is encoded FF ?

These are direct and indirect call, same as there is direct and indirect jump.
The opcodes are:

Near direct jumps are relative, far direct jumps are absolute.
The symmetry between JMP and CALL is broken only by the direct short jump - the JMP instruction with 8-bit displacement (opcode EB).

The closest I've gotten to a direct jump is by using a temporary register. As in
mov eax,0x12345678
jmp eax

It is pretty compact and fast, if I am not mistaken.

I would advise against push xxxxxxx retn combination.

Although I haven't tested it first hand, according to Intel, every call should be matched up with a return. Pentiums have a call/return branch prediction buffer, and returning to an arbitrary location flushes it, causing all kinds of performance problems.

You are correct - the mov eax,location \ jmp eax is better because it can be predicted by that tiny knowledge the CPU has been given.

push modifies the esp and writes over any information at current esp offset. Return again changes esp. Neither of the sequences change flags so overall the mov variant is better. Although the push variant is ALWAYS shorter 1 to 3 bytes is you want that