I'm not sure if you can actually assemble that code. It's not 'illegal' as such, but it basically doesn't make sense.
movs does not modify any flags, so rep is the only sensible instruction to use with it.
repe/repne would either not run at all, or behave like rep, depending on the value of the zero flag. So it will be error-prone to use this instruction.
I suspect that is why certain assemblers will not allow these prefixes on movs.
There is a simple workaround: just emit the prefix byte manually:

1db 0F2h ; repne
2movsw

I suppose you could use it to create conditionally executing lods/stos/movs instructions, assuming it works the way you expect it to.
Instead of:

1cmp ax, bx
2jne skip
3rep movsw
4skip:

You could do:

1cmp ax, bx
2repe movsw

MASM got the /Zm option for maximum Masm 5.1 compatiblity (that worked with former MASM version i think) - its shown using /? but it seems not to work - still the error
USAM got the -Zm option but also do not seems to supress the error

thx for the info: i will use a REPNE_MOVS macro in all places to fix that (and other 100% binary identical breaking stuff)

Note that repe uses the same prefix as rep (0xF3), so the given example wouldn't work as expected.

Regardless the repe/repne prefixes only check the Z flag when the string instruction updates the flags (only the count register is tested before the first execution) so if the flags are never updated only a zero count register can terminate the loop.

so i think replacing repne/repnz on movs with rep would be the correct asm code for masm resulting in the same behavior
but i will stay with db 0f2h, 0a5h for beeing binary equal 😀

Technically replacing F2 A5 with F3 A5 would be correct because the former is undefined, but intel/amd aren't likely to change behaviour of it because it's been mistakenly used in a lot of MBR code.

jmarsh wrote:

Note that repe uses the same prefix as rep (0xF3), so the given example wouldn't work as expected.

You're right.
Which also implies that MASM's behaviour is 'syntactic sugar'... Apparently they want to force pairing rep with movs/lods/stos, and repe/repne with cmps/scas.

thx for the help

Scali wrote:

I'm not sure if you can actually assemble that code. It's not 'illegal' as such, but it basically doesn't make sense.
movs does not modify any flags, so rep is the only sensible instruction to use with it.

It depends very much on one thing: if the Z flag was controlled beforehand.

1cmp dl, 12h
2repne movsw

will perform the copy if DL wasn't 12h. i.e. it's a conditional copy, or a way to update SI and DI without altering flags further.
The same as repz lodsb, for example, to conditionally update SI without altering flags further.

That is incorrect. repe/repne do not check the initial value of the Z flag, it is only checked after the string instruction updates it.
For instructions that don't update the flags, repe and repne act identically (both as plain rep).

jmars seems to be right: https://c9x.me/x86/html/file_module_x86_id_279.html

imo ne/nz is for usage on scas/cmps than movs.