Re: pb with frame notes from hppa_expand_prologue ?

[Olivier Hainque <hainque at act-europe dot fr>]

law@redhat.com writes:
 > I believe you need to set FRAME_RELATED_P on the sub expressions within
 > the PARALLEL.  Something like this:
 > 
 >      /* Set RTX_FRAME_RELATED_P on the insn and each member of the
 >          PARALLEL for the dwarf2 unwinder.  */
 >       RTX_FRAME_RELATED_P (insn) = 1;
 >       for (i = 0; i < XVECLEN (PATTERN (insn), 0); i++)
 >         RTX_FRAME_RELATED_P (XVECEXP (PATTERN (insn), 0, i)) = 1;
 > 
 > Where INSN is the value returned by emit_insn (gen_post_store ...))

 This is actually what I went on with, thanks :)

 I hit a couple of other problems later, and an interesting one had to do with
 the way floating point registers are saved, using scratch r1 as a temporary
 base register :

     /* First get the frame or stack pointer to the start of the FP register
         save area.  */
      if (frame_pointer_needed)
        FRP (set_reg_plus_d (1, FRAME_POINTER_REGNUM, offset));
      else
        FRP (set_reg_plus_d (1, STACK_POINTER_REGNUM, offset));


 A side effect of this is to make r1 the new base register for CFA
 computation, which does not seem to work because r1 is caller-saved and its
 value is not available when unwinding the stack at run-time.

 I temporarily kludged this by modifying rule 4 in dwarf2out_frame_debug_expr
 for the cases in which reg1 is caller-saved, but I think that's not the right
 way to go.

 Any insight as to what should be done for this case ?

 Thanks for your help,

 Olivier