[LLVMdev] Trouble with inline asm

[Chris Lattner]

On Jun 7, 2008, at 11:16 AM, Tomas Lindquist Olsen wrote:

>>
>> Good luck doing this; I imagine this is going to be tricky to
>> implement, especially converting from Intel to AT&T syntax.
>
> Thanx, we'll need it ;) Fortunately much of the code is already in
> place (taken from the GDC D compiler), so it "only" had to be adapted
> from GCC to LLVM.

Incidentally, llvm-gcc has partial support for this with the -fasm- 
blocks command line option.  This code is accepted by apple gcc:

int printf(char*,...);
int main()
{
    int i = 12;
    printf("%d\n", i);
    asm
    {
        mov EAX, i;
        add EAX, EAX;
        mul EAX, 2;
        mov i, EAX; // ***
    }
    printf("%d\n", i);
    return 0;
}

But it looks like it is currently rejected by llvm-gcc (invalid or  
unsupported asm).  This is a bug that we will fix in time.  I think  
that the GCC front-end makes the same mistake as the D front-end,  
where it lowers each statement into a separate inline asm node.   
Here's an example that does work with llvm-gcc, though it produces  
ugly code like Apple GCC (pinning 'i' to the stack):

int printf(char*,...);
int main()
{
    int i = 12;
    printf("%d\n", i);
    asm
    {
        shl i, 1
    }
    printf("%d\n", i);
    return 0;
}


-Chris