Re: [RTL]Access rtx oprands

[王 逸 <cnnjuwy at hotmail dot com>]

In gcc/rtl.h:

/* Holds a unique number for each insn.
   These are not necessarily sequentially increasing.  */
#define INSN_UID(INSN)  XINT (INSN, 0)

/* Chain insns together in sequence.  */
#define PREV_INSN(INSN) XEXP (INSN, 1)
#define NEXT_INSN(INSN) XEXP (INSN, 2)

#define BLOCK_FOR_INSN(INSN) XBBDEF (INSN, 3)
#define INSN_LOCATOR(INSN) XINT (INSN, 4)
/* The body of an insn.  */
#define PATTERN(INSN) XEXP (INSN, 5)

So I think XEXP(insn, 0) gets the insn_uid while XEXP(insn, 5) gets the call rtx in a call_insn, this right?

The if conditions listed in my original message were all true, just the last evaluate statement got a segment fault, while compiling GCC.
I'm confused why 
    (GET_CODE(insn) == SYMBOL_REF)
 is true, but 
    const char *name = XSTR(insn, 0)
araised a segment fault?

thanks

WangYi


----- Original Message ----- 
From: "Dave Korn" <dk@artimi.com>
To: "'王 逸'" <cnnjuwy@hotmail.com>; <gcc@gcc.gnu.org>
Sent: Wednesday, June 23, 2004 5:49 PM
Subject: RE: [RTL]Access rtx oprands


> 
> > -----Original Message-----
> > From: gcc-owner On Behalf Of ? ?
> > Sent: 25 June 2004 21:45
> 
> > Suppose an rtx variable insn points to a call_insn rtx, and 
> > this rtx dumped with "gcc -dr ..." looks like
> > (call_insn 12 11 8 (call (mem:QI (symbol_ref:SI ("my_fun")) [0 S1 A8])
> >         (const_int 16 [0x10])) -1 (nil)
> >     (expr_list:REG_EH_REGION (const_int 0 [0x0])
> >         (nil))
> >     (nil))
> > How could I get the symbol_ref "my_fun" as an string?
> > 
> > I read some gcc source codes and patched gcc with the 
> > following statements:
> >     if(GET_CODE(insn) == CALL_INSN)
> >       if(GET_CODE(XEXP(insn, 5)) == CALL)
> >         if(GET_CODE(XEXP(XEXP(insn, 5), 0)) == MEM)
> >           if(GET_CODE(XEXP(XEXP(XEXP(insn, 5), 0), 0)) == SYMBOL_REF)
> >            //      const char *name = "ok";
> >             const char *name = XSTR(XEXP(XEXP(XEXP(insn, 5), 
> > 0), 0), 0);    
> > 
> > I got an segment fault error when compiling GCC, but if I 
> > replace the evaluate statement with the commented one, then GCC
> > is successfully built and the patched GCC works well.
> > 
> > There are some similar usages of XEXP and XSTR in GCC source, 
> > I wonder why this one doesn't work?
> > If my method of getting the routine name in an call_insn is 
> > wrong, please tell me how to get that.
> 
>   Those three numbers are the UIDs in the insn chain, not operands, and I'm
> still not quite sure how you counted 5.  You want to access XEXP (insn, 0)
> to get at the call rtx; trying to access XEXP (insn, 5) will get you an
> undefined pointer.
> 
> 
>     cheers, 
>       DaveK
> -- 
> Can't think of a witty .sigline today....
> 
>