[Bug fortran/45318] New: Do more parenthesis simplification with -fno-protect-parens

[Bug fortran/45318] New: Do more parenthesis simplification with -fno-protect-parens

[burnus at gcc dot gnu dot org]

Fortran preserves () in expressions - in many cases, it shouldn't need to do so
if 
-fno-protect-parens is specified. The option currently affects only the middle
end, but there might be some cases where it is preserved without needing to be.
(Though, I might be wrong and everything is already properly simplified).

Note: The out most parenthesis can not always be removed without changing the
semantics, e.g.

  call copy ( (a), a)
contains
  subroutine copy (x, y)
    intent(in) :: x
    intent(out) :: y
is valid but only works if the "(a)" [-> temporary] is not optimized to "a".

In some cases, one might need to check for the unsave_math_optimization flag
before changing, e.g., "2+(a-2)" to "a" - or rather "(a)".


-- 
           Summary: Do more parenthesis simplification with -fno-protect-
                    parens
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: burnus at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45318

[Bug fortran/45318] Do more parenthesis simplification with -fno-protect-parens

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2010-08-18 14:23 -------
(In reply to comment #0)
> In some cases, one might need to check for the unsave_math_optimization flag
> before changing, e.g., "2+(a-2)" to "a" - or rather "(a)".

The whole point of PAREN_EXPR in the middle-end is to avoid transforming
2 + (a - 2) even _with_ -funsafe-math-optimizations / -ffast-math!


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45318

[Bug fortran/45318] Do more parenthesis simplification with -fno-protect-parens

[burnus at gcc dot gnu dot org]

------- Comment #2 from burnus at gcc dot gnu dot org  2010-08-18 17:44 -------
(In reply to comment #1)
> (In reply to comment #0)
> > In some cases, one might need to check for the unsave_math_optimization flag
> > before changing, e.g., "2+(a-2)" to "a" - or rather "(a)".
> 
> The whole point of PAREN_EXPR in the middle-end is to avoid transforming
> 2 + (a - 2) even _with_ -funsafe-math-optimizations / -ffast-math!

Well, I am talking about FE optimizations with -fno-protect-parens (note the
"no-"). The PAREN_EXPR protection *is* and *should* one done with
-fprotect-parens (which is the default setting).

In terms of the middle end, -fno-protect-parens and -fprotect-parens are both
properly handled.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45318

[Bug fortran/45318] Do more parenthesis simplification with -fno-protect-parens

[dfranke at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45318

Daniel Franke <dfranke at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2010.12.28 23:37:54
                 CC|                            |dfranke at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #3 from Daniel Franke <dfranke at gcc dot gnu.org> 2010-12-28 23:37:54 UTC ---
Given Richard's comments and no testcase, I'm in doubt if this PR is
meaningful?!

[Bug fortran/45318] Do more parenthesis simplification with -fno-protect-parens

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45318

--- Comment #4 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> Given Richard's comments and no testcase, I'm in doubt if this PR 
> is meaningful?!

Any new opinion about this PR? or should we close it as WONTFIX.

NOTE: I'ld like very much that the -no-protect-parens be removed from -Ofast!-(

[Bug fortran/45318] Do more parenthesis simplification with -fno-protect-parens

[tkoenig at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45318

Thomas Koenig <tkoenig at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tkoenig at gcc dot gnu.org

--- Comment #5 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
Any specific examples?

[Bug fortran/45318] Do more parenthesis simplification with -fno-protect-parens

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45318

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|---                         |WORKSFORME

--- Comment #6 from Tobias Burnus <burnus at gcc dot gnu.org> ---
(In reply to Thomas Koenig from comment #5)
> Any specific examples?

I looked now through the code (for INTRINSIC_PARENTHESES) and glanced at some
of the related PRs. I failed to find a case where one could do something in the
FE - and the ME already handles it. (An example would be "(a+5)-7" - however,
the ME essentially ignores all code involving nonconstants; the only exception
is dependency tracing and there such a case is also unlikely.)

In summary, while there could be in principle a case, which should be handled,
in practice, there does not seem to be one.

Thus: Close as WORKSFORME.