[LLVMdev] Instructions having variable names as operands

[Chris Lattner]

On Wed, 20 Dec 2006, Seung Jae Lee wrote:
> Dear Mr. Lattner:
> Thank you for kind information.
> I am still a little confused, though.
>
> In your example,
>   %X = add int %Y, %Z
> is generated to
>   add EAX, EBX
>
> I think EAX and EBX are the register names of X86.
> But, I should emit variable names instead of register names.
>
> For my example in the source code,
>    int k;
>    k = i + j;
> should be emit like this:
>    reg k
>    add i,j;k
> (If register names are still used, this can be shown like this:
>    reg EAX
>    add EAX,EBX;EAX
> Isn't it?
> )
> Would you mind telling me w.r.t this?
> Thank you very much.

How does your "instruction set" handle code like:

int foo() {
   int X = 1;  // X1
   if (somecond) {
     int X = 2;  // X2

     return X;
   }
   return X;
}

Surely you don't have full source language scoping rules.  If you want 
this, you need to look at debug info.

-Chris

-- 
http://nondot.org/sabre/
http://llvm.org/