Richard Sandiford via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> How about instead:
>
> (1) Define a new ASLK_* flag for assign_stack_local_1.
>
> (2) When the flag is set, make:
>
> if (alignment_in_bits > MAX_SUPPORTED_STACK_ALIGNMENT)
> {
> alignment_in_bits = MAX_SUPPORTED_STACK_ALIGNMENT;
> alignment = MAX_SUPPORTED_STACK_ALIGNMENT / BITS_PER_UNIT;
> }
>
> increase the size by (new_align - old_align).
Er, I meant old_align - new_align here :-)
>
> (3) When the flag is set, and the new alignment is smaller than the
> original alignment, call align_dynamic_address before creating
> the MEM, as in your original patch.
>
> There are probably other details, but it looks like that should cope
> with the x86 dynamic stack realignment scheme even with the max-based
> alignment calculation discussed above.
>
> Thanks,
> Richard
