On Sun, Jun 24, 2012 at 3:36 AM, Richard Sandiford
<rdsandif...@googlemail.com> wrote:
> One of the more unfortunate things MIPS has inherited is an LP64 ABI
> that uses 32-bit rather than 64-bit ELF. I've no idea how many people
> use it these days (if anyone), but it happens to the ABI of the MIPS
> primary target, mipsisa64-elf.
>
> Because we're using 32-bit ELF, DWARF2_ADDR_SIZE is 4 rather than 8.
> This currently causes problems with dwarf2out when we have an eliminated
> expression such arg_pointer_rtx or arg_pointer_rtx + CONST. Normally we'd
> treat them specially and apply the elimination, but this implicitly
> requires Pmode to be no bigger than DWARF2_ADDR_SIZE:
>
> case REG:
> if (GET_MODE_CLASS (mode) != MODE_INT
> || (GET_MODE_SIZE (mode) > DWARF2_ADDR_SIZE
> #ifdef POINTERS_EXTEND_UNSIGNED
> && (mode != Pmode || mem_mode == VOIDmode)
> #endif
> ))
> {
> [...]
> mem_loc_result = new_loc_descr (DW_OP_GNU_regval_type,
> dbx_reg_number (rtl), 0);
> [...]
> }
> if (REGNO (rtl) < FIRST_PSEUDO_REGISTER)
> mem_loc_result = based_loc_descr (rtl, 0, VAR_INIT_STATUS_INITIALIZED);
> [...]
>
> case PLUS:
> plus:
> if (is_based_loc (rtl)
> && GET_MODE_SIZE (mode) <= DWARF2_ADDR_SIZE
> && GET_MODE_CLASS (mode) == MODE_INT)
> mem_loc_result = based_loc_descr (XEXP (rtl, 0),
> INTVAL (XEXP (rtl, 1)),
> VAR_INIT_STATUS_INITIALIZED);
>
> where based_loc_descr does the elimination.
>
> So as things stand we take the first "if", and try to get the register
> number of the fake arg_pointer_rtx. This was latent for a while
> (at least in the sense of GCC testing, although probably not if
> you actually tried to debug using it) until H.J.'s recent patch
> to add an assert to dbx_reg_regnumber. Which just goes to show
> how good an assert it is.
I added it for
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52857
> I realise it's a bit clunky, but is the attached patch OK?
> Tested on mipsisa64-elf, where it restores the build, and on
> x86_64-linux-gnu.
>
> Richard
>
>
> gcc/
> * dwarf2out.c (mem_loc_descriptor): Ignore DWARF2_ADDR_SIZE
> for frame- and argument-pointer reg+offset expressions.
>
It looks the same as
http://gcc.gnu.org/ml/gcc-patches/2012-04/msg01815.html
and my patch has 2 testcases.
Thanks.
--
H.J.
