On May 6, 2014, at 8:07 AM, Nicholas Clifton <ni...@redhat.com> wrote: > + tmp = size_int ((5 * MAX (POINTER_SIZE, BITS_PER_WORD)) / > BITS_PER_WORD - 1);
This is not right. The denominator should either be GET_MODE_SIZE (Pmode) or POINTER_SIZE. See get_nl_goto_field for additional fun. Also, the save area for the machine is exactly 2 * GET_MODE_SIZE (Pmode) in and the save area is exactly GET_MODE_SIZE (STACK_SAVEAREA_MODE (SAVE_NONLOCAL)) bits. builtin_setjmp_setup is the only unaccounted for component. mips is alone in having slop not accounted for in GET_MODE_SIZE (STACK_SAVEAREA_MODE (SAVE_NONLOCAL)). Personally, I’d rather make the size calculation exact, rather than even more sloppy and hard to comprehend. How about GET_MODE_SIZE (STACK_SAVEAREA_MODE (SAVE_NONLOCAL)) / GET_MODE_SIZE (Pmode) + 2 + /* slop for mips, see builtin_setjmp_setup */ 1 - 1. This retains the slop for mips, and fixes ports like ia64 and s390 (see STACK_SAVEAREA_MODE on those ports, it is larger one might expect)? Last patch that `broke’ this: http://gcc.gnu.org/ml/gcc-patches/2004-01/msg02000.html That entire thread is interesting.
