> I think re-test is just waste of cycles, as you know cfa_base_rtx is
> arg_pointer_rtx on both x86_64 and i686 (it is frame_pointer_rtx only on
> pa, vax and rx).  IMHO just testing if var-tracking still compiles without
> warnings ought to be enough.

Alright, quickstrapped and guality-regtested only.  Note that I tweaked the 
test as it didn't FAIL for me (it was UNSUPPORTED on i586 because of a "cannot 
access memory" error from within GDB and PASSed on x86-64).

Thanks for your help.


2011-05-30  Jakub Jelinek  <jakub@redhat.com>
            Eric Botcazou  <ebotcazou@adacore.com>

        * var-tracking.c (vt_add_function_parameter): Remap incoming MEMs with
        crtl->args.internal_arg_pointer based address to arg_pointer_rtx if
        there is a DRAP register and arg_pointer_rtx is the CFA pointer.
        (vt_init_cfa_base): Don't equate cfa_base_rtx if stack was realigned.
        (vt_initialize): Initialize cfa_base_rtx if there is a DRAP register.


2011-05-30  Jakub Jelinek  <jakub@redhat.com>
            Eric Botcazou  <ebotcazou@adacore.com>

	* gcc.dg/guality/drap.c: New test.


-- 
Eric Botcazou