[Bug middle-end/112877] New: TARGET_PROMOTE_PROTOTYPES is not honored consistently, should maybe not apply to builtins

[Bug middle-end/112877] New: TARGET_PROMOTE_PROTOTYPES is not honored consistently, should maybe not apply to builtins

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112877

            Bug ID: 112877
           Summary: TARGET_PROMOTE_PROTOTYPES is not honored consistently,
                    should maybe not apply to builtins
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rguenth at gcc dot gnu.org
  Target Milestone: ---

Currently only the C, C++ and Ada frontends honor (maybe partly)
TARGET_PROMOTE_PROTOTYPES so this is at most an optimization feature,
in particular it doesn't look like the target can guarantee ABI constraints
with this (zero/sign extending arguments at callers).

TARGET_PROMOTE_PROTOTYPES should be (and is AFAICS) reflected into
DECL_ARG_TYPE.

combine is the only thing lookig at DECL_ARG_TYPE.

The iq2000, microblaze and PA targets look at DECL_ARG_TYPE (didn't look
at what).


In particular the vectorizer is pessimized by the promotion when
vectorizing builtins working on < int arguments and for OMP SIMD
function vectorization.  Targets might get more control when
the hook were passed a function decl rather than only a function type.

But since nothing on the GIMPLE level cares about DECL_ARG_TYPE
it might be possible to reflect TARGET_PROMOTE_PROTOTYPES only during
RTL expansion.

[Bug middle-end/112877] TARGET_PROMOTE_PROTOTYPES is not honored consistently, should maybe not apply to builtins

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112877

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2023-12-09

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
. See PR 48274 also.