[Bug tree-optimization/18374] New: Argument and return value marshalling at tree level

[Bug tree-optimization/18374] New: Argument and return value marshalling at tree level

[nathan at gcc dot gnu dot org]

This is a catch-all report to mention that we should really be doing ABI
mandated argument and return value marshalling at the tree level. These
operations can involve sign/zero extension (at the very least) and should be
exposed to the tree optimizers.

-- 
           Summary: Argument and return value marshalling at tree level
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: nathan at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18374

[Bug tree-optimization/18374] Argument and return value marshalling at tree level

[nathan at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|                            |16797
              nThis|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18374

[Bug tree-optimization/18374] Argument and return value marshalling at tree level

[nathan at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|                            |16802
              nThis|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18374

[Bug tree-optimization/18374] [meta-bug] Argument and return value marshalling at tree level

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
           Keywords|                            |meta-bug, missed-
                   |                            |optimization
            Summary|Argument and return value   |[meta-bug] Argument and
                   |marshalling at tree level   |return value marshalling at
                   |                            |tree level


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18374

[Bug tree-optimization/18374] [meta-bug] Argument and return value marshalling at tree level

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-11-17 13:13 -------
Here is a bug which says the oposite should be done (as arguments are already done): PR 15484.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18374

[Bug tree-optimization/18374] [meta-bug] Argument and return value marshalling at tree level

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-11-17 15:53:28
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18374

[Bug tree-optimization/18374] [meta-bug] Argument and return value marshalling at tree level

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18374

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> 2013-02-26 11:33:15 UTC ---
On the other hand we do have a mismatch between caller and callee, for C at
least:

short barshort (short a)
{
  return a;
}
short foo1short (short a)
{
  return barshort(a);
}


;; Function barshort (null)
;; enabled by -tree-original

{
  return a;
}

;; Function foo1short (null)
;; enabled by -tree-original

{
  return barshort ((int) a);
}


this at least makes us need to "fixup" arguments during inlining which
should not be necessary with a consistent IL (also see the DECL_ARG_TYPE
wart we have ... in this case it says 'int' for 'a').

[Bug tree-optimization/18374] [meta-bug] Argument and return value marshalling at tree level

[Paul Schlie]

> Additional Comments From pinskia at gcc dot gnu dot org  2004-11-17
> Here is a bug which says the oposite should be done (as arguments are
> already > done): PR 15484.

Unless I misunderstand; C's promotion/evaluation semantics need not
be literally applied, but must only yield equivalent results to as if
they were. 

Thereby providing opportunities to the compiler to avoid the otherwise
needless promotion of expression/argument precision if it's known that
their use in an alternate form will not affect a resulting utilized
variable or control flow state.

Thereby enabling the compiler to avoid the promotion of arguments to
functions, and correspondingly only require their promotion within
functions if required.

If an ABI specifies that this may not be done, then that's that; but
if it's allowed, or not specified, the compiler should be enabled to
perform this type of global optimization, I would think. (although
GCC may not presently takes full advantage of this potential today.)