OSDev.org

Ok. I've just decided to try learning GNU as using Intel syntax, but I'm running into problems. And I can't even seem to find much documentation related to my problems, and some of the documentation I can find, seems to be wrong.

I'm using this directive: The first problem I ran into is with this line of code: Now I would expect that esp would then point to the same address as __stack_end, but that's not what happens. It gets translated into a reference to the memory at __stack_end, so NASM disassembles it like this: Well, I replaced it with an lea instruction so I still get the behavior I want, but I want to know how to do it with a mov instruction. And why do they even do that? Treating a bare label as a memory reference in this context is inconsistent, as a bare label refers merely to the address in other contexts.

The second problem I ran into is a warning for this code: or even this code: The warning is: "Warning: Treating `dword ptr [0xb8000]' as memory reference"
That doesn't make sense. Of course it's treating it as a memory reference. I very explicitly told it that it's a memory reference. Why would it give me a warning here where I explicitly said it's a memory reference and not give me a warning with my other code, where intuitively it should not have been a memory reference (and isn't in other assemblers)?
So how do I get rid of this warning? Is my syntax wrong? If so, how do I do it right?

(By the way, yes I know moving eax to address 0xb8000 (in the way I'm doing it) looks a little strange, but it's just temporary code to display two characters on the screen.)

I remember GAS always expecting a $ before anything that is to be treated as a value, not a memory reference. Might want to try that.

JohnnyTheDon wrote:I remember GAS always expecting a $ before anything that is to be treated as a value, not a memory reference. Might want to try that.

Yeah, I tried that before. Here are some of the things I've tried and the results: Result: Treats `$__stack_end' as a different symbol than `__stack_end'.
linker says: undefined reference to `$__stack_end'
Result: Probably treats `$' as a symbol.
assembler says: Error: invalid operand for 'mov' ('__stack_end' unexpected)
Result: Treats `__stack_end' as a memory reference, just like my original problem.

Edit: Thanks for the suggestion though.

gas also uses the syntax: op src, dest

so for example the code in intel syntax:

mov eax, 5

would be this in AT&T syntax:

mov $0x5, %eax

would put 5 into eax.

so to fix your problem use
mov $__stack_end, %esp

you should really learn AT&T syntax before you use it.

and you don't need to use .intel_syntax prefix

The whole point was that the OP is trying to use gas to assemble code that uses intel syntax. Thats why '.intel_syntax prefix' is there.

iammisc wrote:gas also uses the syntax: op src, dest

so for example the code in intel syntax:

mov eax, 5

would be this in AT&T syntax:

mov $0x5, %eax

would put 5 into eax.

so to fix your problem use
mov $__stack_end, %esp

you should really learn AT&T syntax before you use it.

and you don't need to use .intel_syntax prefix

I know AT&T is it's native syntax. That's precisely the reason I've (mostly) avoided using Gas all these years: because AT&T syntax is backwards from all the instruction set documentation (except when it's not, which makes it even more confusing). In other words, I'm not going to use AT&T syntax unless I absolutely have to; which I don't, because there are other assemblers. I just hoped that I could stick with a single tool set.

As it turns out the label issue is more annoying than I thought if I can't get it to work right, because it's also a problem with this statement: Naturally, I want to push the address of the string, but instead it pushes the first four characters of the string as a dword. That would be quite annoying if I always have to replace label pushes with this: Is there anyone out there that uses Intel syntax on Gas that could help me?

Well, I figured gcc had to push label addresses sometimes, so I wrote a simple program that required it, and had gcc output intel syntax assembly instead of object code. It turns out gcc does it like this (with a mov instead of a push, but that's not important): After a little web searching, I think the word "FLAT" and the colon are unnecessary, but I'm not 100% sure. It did work fine in my code without them.
Now if only I can figure out what the warning in my first post is about.

You might want to try yasm or nasm. They do intel syntax and IMHO are much better than gas.

He said he wants to learn GNU as with Intel syntax, we don't know why so we have to assume he's got his reasons. Suggesting a different tool doesn't solve his problems...

Just to be sure: What version of GNU as are you using?

I suppose to make reasonable code, you would have to make hackish db statements with hex codes.. gas is evil...

I don't suppose you'll explain why you aren't going with an assembler that has a syntax you recognize?

I'm guessing it's because he wants to keep things consistent, not wanting to use a different syntax for inline assembly than for pure assembly files.

JohnnyTheDon wrote:You might want to try yasm or nasm. They do intel syntax and IMHO are much better than gas.

nekros wrote:I don't suppose you'll explain why you aren't going with an assembler that has a syntax you recognize?

I've been programming in nasm since before yasm existed, but I've also been using gcc all that time for my C code, which means my inline assembly syntax was quite different than my regular assembly syntax, and it's always bothered me. Plus there are a few things about gas syntax that I DO like better. I just couldn't ever get past the reversed operands of AT&T syntax. So I figured I would experiment using gas/gcc with intel syntax in both. If I don't like it, I'll go back to nasm/gcc (but maybe with intel syntax in gcc), or maybe switch to yasm/gcc (which as I understand is basically the same as nasm/gcc). As of right now, it's still up in the air.

Although, if you know of a good, free (as in water, because I don't drink beer) C compiler for Linux that supports Intel-syntax inline assembly, I might try that. With the exception of the Intel compiler, that is, because I couldn't convince it to install on my current Linux distro (although it could be that I just need to try to be a little more convincing).

Solar wrote:He said he wants to learn GNU as with Intel syntax, we don't know why so we have to assume he's got his reasons. Suggesting a different tool doesn't solve his problems...

Just to be sure: What version of GNU as are you using?

as: "GNU assembler version 2.17.50.0.18-1 20070731"
gcc: "gcc (GCC) 4.1.2 20070925 (Red Hat 4.1.2-33)"

JohnnyTheDon wrote:The whole point was that the OP is trying to use gas to assemble code that uses intel syntax. Thats why '.intel_syntax prefix' is there.

Ah, ok. sorry.

Well then, he doesn't need to use percent signs OR dollar signs or any of that AT&T stuff. Just use regular intel syntax.

iammisc wrote:

JohnnyTheDon wrote:The whole point was that the OP is trying to use gas to assemble code that uses intel syntax. Thats why '.intel_syntax prefix' is there.

Ah, ok. sorry.

Well then, he doesn't need to use percent signs OR dollar signs or any of that AT&T stuff. Just use regular intel syntax.

I'm not sure if it matters, but when I did gas with intel syntax in my OS, I did I'm not sure of the difference though