IA-64 issue H-1: Return address register number

[dehnert at baalbek dot engr dot sgi dot com (Jim Dehnert)]

> From davea@quasar  Mon Apr  3 16:50:57 2000
> 
> Jim, note that there is a restriction in dwarf
> in .debug_frame that means the
>
> 	dwarf number
> 
> for the return address register should be <= 255 (fit in
> an unsigned byte).
> 
> This is of little practical importance since .debug_frame is
> not being used, but for a *dwarf* numbering it seems
> silly to make it impossible to create .debug_frame at all.

We have an issue here with our register numbering, which assigns 1-127
to the GRs and 128-255 to the FRs, leaving nothing in the one-byte
range for the return address "register" (which is not an actual
register but an abstract register).  We have three choices that I
see:

 1) Ignore it.  Noone currently intends to use debug_frame for
    IA-64, as we are using the ABI unwind info instead.  See (3).

 2) Shift our numbering of the registers to leave a hole for the
    return address register, e.g. move the FRs to 129+ (or higher).

 3) The problem occurs in the CIE header for debug_frame, where the
    return address is currently specified as a ubyte.  The CIE also has
    a version number, so we could bump the version number and use a
    ULEB for the return address field.  That change has been suggested
    as a generic DWARF change, but I imagine IA-64 implementations
    could choose to do it in any case.

(1) is easiest today, but has a minor chance of causing confusion in
the future if debug_frame implementations arise.  (2) is cleanest,
but might be a minor compatibility problem with already-existing IA-64
implementations.  (3) is OK if the base DWARF group adopts it anyway,
but might be problematic if they don't.

What are your thoughts, especially the current IA-64 implementors?

Jim

-		Jim Dehnert  x3-4272