[Bug middle-end/36106] #pragma omp atomic issues with floating point types

[Bug middle-end/36106] #pragma omp atomic issues with floating point types

[jakub at gcc dot gnu dot org]

------- Comment #1 from jakub at gcc dot gnu dot org  2008-05-02 11:40 -------
Created an attachment (id=15562)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15562&action=view)
gcc44-pr36106-test.patch

Testcases.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106

[Bug middle-end/36106] New: #pragma omp atomic issues with floating point types

[jakub at gcc dot gnu dot org]

Attached testcases fail on x86_64 (with cmpxchg16b) or i386 (with cmpxchg8b),
in both cases the problem is that reading the original value using FPU and
storing it into a temporary doesn't mean bitwise identical copy (in the first
case because long double is said to be a 16 byte type, even when the copy
through FPU actually copies just 80 bits, and in the latter case because the
copy turns the signalling NaN into a corresponding quiet NaN).  This means both
testcases hang.
Will work on a fix.


-- 
           Summary: #pragma omp atomic issues with floating point types
           Product: gcc
           Version: 4.3.1
            Status: UNCONFIRMED
          Keywords: openmp
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jakub at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106

[Bug middle-end/36106] #pragma omp atomic issues with floating point types

[jakub at gcc dot gnu dot org]

------- Comment #2 from jakub at gcc dot gnu dot org  2008-05-05 10:39 -------
This is nothing target specific though, on many targets reading a float or
double sNaN into a FPU register and storing back somewhere else leads to qNaN
being stored.


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |middle-end


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106

[Bug target/36106] #pragma omp atomic issues with floating point types

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2008-05-05 15:40 -------
That sounds like a bug in the target back-ends for not storing exactly the
bit-wise for NaNs.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |target


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106

[Bug middle-end/36106] #pragma omp atomic issues with floating point types

[jakub at gcc dot gnu dot org]

------- Comment #4 from jakub at gcc dot gnu dot org  2008-05-05 15:47 -------
Not really, as IEEE754 mandates turning sNaN into qNaN after raising the
exception.  I have a fix in the works in omp-low.c (so far works for non-SSA
and not yet for -ftree-parallelize-loops) and that's middle-end.


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |middle-end


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106

[Bug middle-end/36106] #pragma omp atomic issues with floating point types

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |jakub at gcc dot gnu dot org
                   |dot org                     |
                URL|                            |http://gcc.gnu.org/ml/gcc-
                   |                            |patches/2008-
                   |                            |05/msg00260.html
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-05-06 09:20:48
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106

[Bug middle-end/36106] #pragma omp atomic issues with floating point types

[jakub at gcc dot gnu dot org]

------- Comment #5 from jakub at gcc dot gnu dot org  2008-05-07 07:28 -------
Subject: Bug 36106

Author: jakub
Date: Wed May  7 07:28:14 2008
New Revision: 135027

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=135027
Log:
        PR middle-end/36106
        * omp-low.c (expand_omp_atomic_pipeline): Load value using the
        integral type rather than floating point, then VIEW_CONVERT_EXPR
        to the floating point type.

        * testsuite/libgomp.c/atomic-5.c: New test.
        * testsuite/libgomp.c/atomic-6.c: New test.
        * testsuite/libgomp.c/autopar-1.c: New test.

Added:
    trunk/libgomp/testsuite/libgomp.c/atomic-5.c
    trunk/libgomp/testsuite/libgomp.c/atomic-6.c
    trunk/libgomp/testsuite/libgomp.c/autopar-1.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/omp-low.c
    trunk/libgomp/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106

[Bug middle-end/36106] #pragma omp atomic issues with floating point types

[jakub at gcc dot gnu dot org]

------- Comment #6 from jakub at gcc dot gnu dot org  2008-05-07 07:56 -------
Subject: Bug 36106

Author: jakub
Date: Wed May  7 07:55:21 2008
New Revision: 135030

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=135030
Log:
        PR middle-end/36106
        * omp-low.c (expand_omp_atomic_pipeline): Load value using the
        integral type rather than floating point, then VIEW_CONVERT_EXPR
        to the floating point type.

        * testsuite/libgomp.c/atomic-5.c: New test.
        * testsuite/libgomp.c/atomic-6.c: New test.
        * testsuite/libgomp.c/autopar-1.c: New test.

Added:
    branches/gcc-4_3-branch/libgomp/testsuite/libgomp.c/atomic-5.c
    branches/gcc-4_3-branch/libgomp/testsuite/libgomp.c/atomic-6.c
    branches/gcc-4_3-branch/libgomp/testsuite/libgomp.c/autopar-1.c
Modified:
    branches/gcc-4_3-branch/gcc/ChangeLog
    branches/gcc-4_3-branch/gcc/omp-low.c
    branches/gcc-4_3-branch/libgomp/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106

[Bug middle-end/36106] #pragma omp atomic issues with floating point types

[jakub at gcc dot gnu dot org]

------- Comment #7 from jakub at gcc dot gnu dot org  2008-05-07 08:05 -------
Fixed.


-- 

jakub at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36106