[Bug other/49914] New: call to abs(long long) in gcc/fold-const.c

[Bug other/49914] New: call to abs(long long) in gcc/fold-const.c

[marc.glisse at normalesup dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

           Summary: call to abs(long long) in gcc/fold-const.c
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: other
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: marc.glisse@normalesup.org


Hello,

in gcc/fold-const.c (currently line 7051), in function
fold_plusminus_mult_expr:

if (exact_log2 (abs (int11)) > 0 && int01 % int11 == 0

where int11 is a HOST_WIDE_INT. Is this call to abs safe? A few lines above,
abs is computed as x>=0?x:-x.

(don't know how to classify this, but it causes a build failure for me with a
C++98 compiler (ambiguous))

[Bug other/49914] call to abs(long long) in gcc/fold-const.c

[marc.glisse at normalesup dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

--- Comment #1 from Marc Glisse <marc.glisse at normalesup dot org> 2011-07-30 18:29:06 UTC ---
Also occurs at gcc/tree-ssa-loop-prefetch.c:797 and in many places in
gcc/tree-ssa-math-opts.c.

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.07.31 09:46:21
          Component|other                       |bootstrap
     Ever Confirmed|0                           |1

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-31 09:46:21 UTC ---
It's not strictly safe.

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |paolo.carlini at oracle dot
                   |                            |com, rguenth at gcc dot
                   |                            |gnu.org

--- Comment #3 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-08-01 16:36:54 UTC ---
So, Richard, what shall we do here? Add to tree.h just:

  #define ABS(X) ((X >= 0) ? (X) : (-X))

and use it in all those places? Or a static inline? Better name?

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[joseph at codesourcery dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

--- Comment #4 from joseph at codesourcery dot com <joseph at codesourcery dot com> 2011-08-01 16:44:52 UTC ---
For HOST_WIDE_INT we want abs_hwi from 
<http://gcc.gnu.org/ml/gcc-patches/2011-07/msg02628.html>.

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

--- Comment #5 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-08-01 16:56:33 UTC ---
Ah, excellent. But what happened to that patch, I see it approved but not
committed!?

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

--- Comment #6 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-08-01 17:20:57 UTC ---
To be clear: if nobody objects I'm willing to bootstrap and test again that
patch, commit it, and then use abs_hwi in all the places pointed out here.

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
                 CC|paolo.carlini at oracle dot |
                   |com                         |
         AssignedTo|unassigned at gcc dot       |paolo.carlini at oracle dot
                   |gnu.org                     |com
   Target Milestone|---                         |4.7.0

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[rguenther at suse dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

--- Comment #7 from rguenther at suse dot de <rguenther at suse dot de> 2011-08-02 08:40:24 UTC ---
On Mon, 1 Aug 2011, paolo.carlini at oracle dot com wrote:

> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914
> 
> Paolo Carlini <paolo.carlini at oracle dot com> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |paolo.carlini at oracle dot
>                    |                            |com, rguenth at gcc dot
>                    |                            |gnu.org
> 
> --- Comment #3 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-08-01 16:36:54 UTC ---
> So, Richard, what shall we do here? Add to tree.h just:
> 
>   #define ABS(X) ((X >= 0) ? (X) : (-X))
> 
> and use it in all those places? Or a static inline? Better name?

No, I'd say have a abs_hwi in hwint.h and use that (ISTR sebastian pop
adding one in a (pending?) patch)

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[rguenther at suse dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

--- Comment #8 from rguenther at suse dot de <rguenther at suse dot de> 2011-08-02 08:42:00 UTC ---
On Mon, 1 Aug 2011, paolo.carlini at oracle dot com wrote:

> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914
> 
> --- Comment #6 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-08-01 17:20:57 UTC ---
> To be clear: if nobody objects I'm willing to bootstrap and test again that
> patch, commit it, and then use abs_hwi in all the places pointed out here.

well, I don't think we want the assert in abs_hwi as it stands now.
Either we can handle this case in the callers or we should just do
what we can - exploit the undefined behavior and return INT_MIN for
INT_MIN.

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[joseph at codesourcery dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

--- Comment #9 from joseph at codesourcery dot com <joseph at codesourcery dot com> 2011-08-02 12:32:24 UTC ---
On Tue, 2 Aug 2011, rguenther at suse dot de wrote:

> well, I don't think we want the assert in abs_hwi as it stands now.
> Either we can handle this case in the callers or we should just do
> what we can - exploit the undefined behavior and return INT_MIN for
> INT_MIN.

If a caller is prepared to get a negative value back from abs_hwi, then an 
abs_hwi_wrap version would seem appropriate for such a caller (or a 
version that returns unsigned HOST_WIDE_INT, if the callers are written to 
work with unsigned values).  If a caller expects the result to be positive 
and always the absolute value (but of a signed type), it's a bug in the 
caller to pass the least HOST_WIDE_INT value, and the assert serves to 
detect such a bug in the caller.

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[paolo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

--- Comment #10 from paolo at gcc dot gnu.org <paolo at gcc dot gnu.org> 2011-08-02 20:27:00 UTC ---
Author: paolo
Date: Tue Aug  2 20:26:57 2011
New Revision: 177212

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=177212
Log:
2011-08-02  Paolo Carlini  <paolo.carlini@oracle.com>

    PR bootstrap/49914
    * fold-const.c (fold_plusminus_mult_expr): Use abs_hwi instead
    of abs.
    * tree-ssa-math-opts.c (gimple_expand_builtin_pow): Likewise.
    * tree-ssa-loop-prefetch.c (prune_ref_by_group_reuse): Likewise.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/fold-const.c
    trunk/gcc/tree-ssa-loop-prefetch.c
    trunk/gcc/tree-ssa-math-opts.c

[Bug bootstrap/49914] call to abs(long long) in gcc/fold-const.c

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49914

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED

--- Comment #11 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-08-02 20:31:09 UTC ---
Fixed. I hope it helps a little bit also the __cplusplus issue...