https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91710
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target Milestone|--- |9.3
Summary|unexpected ABI change note |[9/10 Regression]
| |unexpected ABI change note
--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
>I wonder whether this isn't simply because aarch64_function_arg_boundary()
>guards its inform() invocation only by a condition not depending on the
>function inputs.
You are wrong when you say it is not dependent on function inputs:
unsigned int alignment = aarch64_function_arg_alignment (mode, type,
&abi_break);
if (abi_break & warn_psabi)
inform (input_location, "parameter passing for argument of type "
"%qT changed in GCC 9.1", type);
abi_break is computed by the function call aarch64_function_arg_alignment
which is computes that value value by its two arguments.
Plus the first thing that aarch64_function_arg_alignment does is zeros out
abi_break.
Before aarch64_function_arg_alignment would return the alignment of DECL_ALIGN
but now it returns alginment of DECL_BIT_FIELD_TYPE instead for this case.
The problem is it could change it but it depends on if:
MIN (MAX (alignment, PARM_BOUNDARY), STACK_BOUNDARY);
Would have a different value between the old one and the new one.
That is if new and old alignment are less than or equal to PARM_BOUNDARY (64),
there is no ABI change. But if it is equal to STACK_BOUNDARY (128), then there
is one.
