[pjmlp]

I loved the asm way in PC compilers used to support inline asm.

    asm {
       mov ax, 0x0013
       int 0x10
    }

Can't really like the way clang and gcc asm work, even it it means giving more info to the optimizer.

[groovy2shoes]

Not to mention that you could just use C identifiers in the asm block willy-nilly:

    int foo(int c)
    {
        int r;
        
        asm {
           mov ax, 0x0b00
           int 0x21              /* poll stdin for input   */
           mov ax, offset c      /* get address of `c`     */
           call bar              /* call other C function  */
           mov r, dx             /* save the result in `r` */
        }
        
        return r;
    }

Or something like that — it's been a while. Those were the days! GCC's (and I'm assuming clang's is much the same) inline assembly is a joke in comparison. I remember, when I first encountered it, flipping back-and-forth through the GCC docs looking for how people actually do inline asm with GCC...