Patchwork ORDERED_EXPR in invert_tree_comparison

login
register
mail settings
Submitter Marc Glisse
Date Aug. 2, 2012, 12:48 p.m.
Message ID <alpine.DEB.2.02.1208021438370.7013@stedding.saclay.inria.fr>
Download mbox | patch
Permalink /patch/174749/
State New
Headers show

Comments

Marc Glisse - Aug. 2, 2012, 12:48 p.m.
On Thu, 2 Aug 2012, Richard Guenther wrote:

> On Wed, Aug 1, 2012 at 9:21 PM, Marc Glisse <marc.glisse@inria.fr> wrote:
>> Hello,
>>
>> an opinion on this?
>>
>> (I just noticed: I'll update the list in the comment visible at the top of
>> the patch if this gets in).
>
> It looks ok to me but I am no floating-point expert.  Can you add a testcase?
>
> Ok with that change.

Here again with a testcase. The -O is not necessary for the optimization 
to happen, but it seemed wrong to me not to include it. I wondered about 
adding an explicit -ftrapping-math, for documentation purposes.

I am redoing the bootstrap+regtest, then I'll commit if I don't hear 
protests about the testcase.

gcc/ChangeLog
2012-06-15  Marc Glisse  <marc.glisse@inria.fr>

 	PR tree-optimization/53805
 	* fold-const.c (invert_tree_comparison): Do invert ORDERED_EXPR and
 	UNORDERED_EXPR for floating point.

gcc/testsuite/ChangeLog
2012-06-15  Marc Glisse  <marc.glisse@inria.fr>

 	PR tree-optimization/53805
 	* gcc.dg/fold-notunord.c: New testcase.
Richard Guenther - Aug. 2, 2012, 1:36 p.m.
On Thu, Aug 2, 2012 at 2:48 PM, Marc Glisse <marc.glisse@inria.fr> wrote:
> On Thu, 2 Aug 2012, Richard Guenther wrote:
>
>> On Wed, Aug 1, 2012 at 9:21 PM, Marc Glisse <marc.glisse@inria.fr> wrote:
>>>
>>> Hello,
>>>
>>> an opinion on this?
>>>
>>> (I just noticed: I'll update the list in the comment visible at the top
>>> of
>>> the patch if this gets in).
>>
>>
>> It looks ok to me but I am no floating-point expert.  Can you add a
>> testcase?
>>
>> Ok with that change.
>
>
> Here again with a testcase. The -O is not necessary for the optimization to
> happen, but it seemed wrong to me not to include it. I wondered about adding
> an explicit -ftrapping-math, for documentation purposes.
>
> I am redoing the bootstrap+regtest, then I'll commit if I don't hear
> protests about the testcase.

Yes, an explicit -ftrapping-math would be good.

Thanks,
Richard.

> gcc/ChangeLog
>
> 2012-06-15  Marc Glisse  <marc.glisse@inria.fr>
>
>         PR tree-optimization/53805
>         * fold-const.c (invert_tree_comparison): Do invert ORDERED_EXPR and
>         UNORDERED_EXPR for floating point.
>
> gcc/testsuite/ChangeLog
>
> 2012-06-15  Marc Glisse  <marc.glisse@inria.fr>
>
>         PR tree-optimization/53805
>         * gcc.dg/fold-notunord.c: New testcase.
>
> --
> Marc Glisse
> Index: gcc/testsuite/gcc.dg/fold-notunord.c
> ===================================================================
> --- gcc/testsuite/gcc.dg/fold-notunord.c        (revision 0)
> +++ gcc/testsuite/gcc.dg/fold-notunord.c        (revision 0)
> @@ -0,0 +1,10 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O -fdump-tree-optimized" } */
> +
> +int f (double d)
> +{
> +  return !__builtin_isnan (d);
> +}
> +
> +/* { dg-final { scan-tree-dump " ord " "optimized" } } */
> +/* { dg-final { cleanup-tree-dump "optimized" } } */
>
> Property changes on: gcc/testsuite/gcc.dg/fold-notunord.c
> ___________________________________________________________________
> Added: svn:eol-style
>    + native
> Added: svn:keywords
>    + Author Date Id Revision URL
>
> Index: gcc/fold-const.c
> ===================================================================
> --- gcc/fold-const.c    (revision 190071)
> +++ gcc/fold-const.c    (working copy)
> @@ -2087,26 +2087,28 @@ static tree
>  pedantic_non_lvalue_loc (location_t loc, tree x)
>  {
>    if (pedantic_lvalues)
>      return non_lvalue_loc (loc, x);
>
>    return protected_set_expr_location_unshare (x, loc);
>  }
>
>  /* Given a tree comparison code, return the code that is the logical
> inverse.
>     It is generally not safe to do this for floating-point comparisons,
> except
> -   for EQ_EXPR and NE_EXPR, so we return ERROR_MARK in this case.  */
> +   for EQ_EXPR, NE_EXPR, ORDERED_EXPR and UNORDERED_EXPR, so we return
> +   ERROR_MARK in this case.  */
>
>  enum tree_code
>  invert_tree_comparison (enum tree_code code, bool honor_nans)
>  {
> -  if (honor_nans && flag_trapping_math && code != EQ_EXPR && code !=
> NE_EXPR)
> +  if (honor_nans && flag_trapping_math && code != EQ_EXPR && code !=
> NE_EXPR
> +      && code != ORDERED_EXPR && code != UNORDERED_EXPR)
>      return ERROR_MARK;
>
>    switch (code)
>      {
>      case EQ_EXPR:
>        return NE_EXPR;
>      case NE_EXPR:
>        return EQ_EXPR;
>      case GT_EXPR:
>        return honor_nans ? UNLE_EXPR : LE_EXPR;
>
Nathan Froyd - Aug. 2, 2012, 1:53 p.m.
On Thu, Aug 02, 2012 at 02:48:08PM +0200, Marc Glisse wrote:
> I am redoing the bootstrap+regtest, then I'll commit if I don't hear
> protests about the testcase.
> 
> gcc/ChangeLog
> 2012-06-15  Marc Glisse  <marc.glisse@inria.fr>
> 
> 	PR tree-optimization/53805
> 	* fold-const.c (invert_tree_comparison): Do invert ORDERED_EXPR and
> 	UNORDERED_EXPR for floating point.

Minor protest about the ChangeLog: I think you mean "Do _not_ invert..."

-Nathan
Marc Glisse - Aug. 2, 2012, 3:20 p.m.
On Thu, 2 Aug 2012, Nathan Froyd wrote:

> On Thu, Aug 02, 2012 at 02:48:08PM +0200, Marc Glisse wrote:
>> I am redoing the bootstrap+regtest, then I'll commit if I don't hear
>> protests about the testcase.
>>
>> gcc/ChangeLog
>> 2012-06-15  Marc Glisse  <marc.glisse@inria.fr>
>>
>> 	PR tree-optimization/53805
>> 	* fold-const.c (invert_tree_comparison): Do invert ORDERED_EXPR and
>> 	UNORDERED_EXPR for floating point.
>
> Minor protest about the ChangeLog: I think you mean "Do _not_ invert..."

No, I do mean do invert. The point of the patch is that even for floating 
point and even with trapping-math, it is still safe to invert them.

Maybe I can reformulate as: "Invert ORDERED_EXPR and UNORDERED_EXPR even 
for trapping floating point." ?
Nathan Froyd - Aug. 2, 2012, 3:22 p.m.
On Thu, Aug 02, 2012 at 05:20:24PM +0200, Marc Glisse wrote:
> On Thu, 2 Aug 2012, Nathan Froyd wrote:
> >>	PR tree-optimization/53805
> >>	* fold-const.c (invert_tree_comparison): Do invert ORDERED_EXPR and
> >>	UNORDERED_EXPR for floating point.
> >
> >Minor protest about the ChangeLog: I think you mean "Do _not_ invert..."
> 
> No, I do mean do invert. The point of the patch is that even for
> floating point and even with trapping-math, it is still safe to
> invert them.

Ahhh, yes.  I misread the patch.

> Maybe I can reformulate as: "Invert ORDERED_EXPR and UNORDERED_EXPR
> even for trapping floating point." ?

That works for me.

-Nathan

Patch

Index: gcc/testsuite/gcc.dg/fold-notunord.c

===================================================================
--- gcc/testsuite/gcc.dg/fold-notunord.c	(revision 0)

+++ gcc/testsuite/gcc.dg/fold-notunord.c	(revision 0)

@@ -0,0 +1,10 @@ 

+/* { dg-do compile } */

+/* { dg-options "-O -fdump-tree-optimized" } */

+

+int f (double d)

+{

+  return !__builtin_isnan (d);

+}

+

+/* { dg-final { scan-tree-dump " ord " "optimized" } } */

+/* { dg-final { cleanup-tree-dump "optimized" } } */


Property changes on: gcc/testsuite/gcc.dg/fold-notunord.c
___________________________________________________________________
Added: svn:eol-style
   + native
Added: svn:keywords
   + Author Date Id Revision URL

Index: gcc/fold-const.c

===================================================================
--- gcc/fold-const.c	(revision 190071)

+++ gcc/fold-const.c	(working copy)

@@ -2087,26 +2087,28 @@  static tree

 pedantic_non_lvalue_loc (location_t loc, tree x)
 {
   if (pedantic_lvalues)
     return non_lvalue_loc (loc, x);
 
   return protected_set_expr_location_unshare (x, loc);
 }
 
 /* Given a tree comparison code, return the code that is the logical inverse.
    It is generally not safe to do this for floating-point comparisons, except
-   for EQ_EXPR and NE_EXPR, so we return ERROR_MARK in this case.  */

+   for EQ_EXPR, NE_EXPR, ORDERED_EXPR and UNORDERED_EXPR, so we return

+   ERROR_MARK in this case.  */

 
 enum tree_code
 invert_tree_comparison (enum tree_code code, bool honor_nans)
 {
-  if (honor_nans && flag_trapping_math && code != EQ_EXPR && code != NE_EXPR)

+  if (honor_nans && flag_trapping_math && code != EQ_EXPR && code != NE_EXPR

+      && code != ORDERED_EXPR && code != UNORDERED_EXPR)

     return ERROR_MARK;
 
   switch (code)
     {
     case EQ_EXPR:
       return NE_EXPR;
     case NE_EXPR:
       return EQ_EXPR;
     case GT_EXPR:
       return honor_nans ? UNLE_EXPR : LE_EXPR;