[Bug fortran/28005] New: gfortran: mathmul produces wrong result

[Bug fortran/28005] New: gfortran: mathmul produces wrong result

[tobias dot burnus at physik dot fu-berlin dot de]

From
http://groups.google.com/group/comp.lang.fortran/browse_frm/thread/cf9a1ed1754c6f4a/ed236efb2be07bc0?tvc=1#ed236efb2be07bc0

The program
------------------------------
 program tmatmul
   implicit none
   integer, parameter         ::  nmax = 3
   integer                    ::  n = 2
   real, dimension(nmax,nmax) ::  B=0.0, C=1.0, X=0.0

   B(1,1) = 1.0
   B(1,2) = 2.0
   B(2,1) = 3.0
   B(2,2) = 5.0

   X(1:n,1) = matmul( B(2,1:n),C(1:n,1:n) )
   print *, X(:,1)
   print *, "  Should be: 8.0  8.0  0.0"
end program tmatmul
------------------------------

produces with gfortran (4.2.0 20060612 and 4.1.0 (SUSE Linux)):

   3.000000       3.000000       0.000000
   Should be: 8.0  8.0  0.0

ifort, g95 and NAG f95 give the correct result (8. 8. 0.0).


-- 
           Summary: gfortran: mathmul produces wrong result
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: tobias dot burnus at physik dot fu-berlin dot de


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] gfortran: mathmul produces wrong result

[pault at gcc dot gnu dot org]

------- Comment #1 from pault at gcc dot gnu dot org  2006-06-13 19:33 -------
Tobias,

Thanks for the recent batch of reports; this one, in particular, which is not
at all good!

I am heavily committed with a few other bugs; I'll see if FX or Thomas have the
time.  Otherwise I'll drop everything for this one.

Paul


-- 

pault at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2006-06-13 19:33:28
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] gfortran: mathmul produces wrong result

[paul dot richard dot thomas at cea dot fr]

------- Comment #2 from paul dot richard dot thomas at cea dot fr  2006-06-14 10:28 -------
Created an attachment (id=11668)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=11668&action=view)
Development fix to the PR

Tobias,

I have fixed the problem for integer_4; I do not have the time to regenerate
the library today but will do so in the next 24 hours, so that the general case
is fixed. I will submit the patch asap afterwards.

The testcase that will be submitted is attached below.

Thanks again

Paul

! { dg-do run }
! Check the fix for PR28005, in which the mechanism for dealing
! with matmul (transpose (a), b) would cause wrong results for
! matmul (a(i, 1:n), b(1:n, 1:n)).
!
! Based on the original testcase contributed by
! Tobias Burnus  <tobias.burnus@physik.fu-berlin.de>
!   
   implicit none
   integer, parameter         ::  nmax = 3
   integer                    ::  i, n = 2
   integer, dimension(nmax,nmax) ::  iB=0 , iC=1
   integer, dimension(nmax,nmax) ::  iX1=99, iX2=99, iChk
   iChk = reshape((/30,66,102,36,81,126,42,96,150/),(/3,3/))

! This would give 3, 3, 99
   iB = reshape((/1 ,3 ,0 ,2 ,5 ,0 ,0 ,0 ,0 /),(/3,3/))
   iX1(1:n,1) = matmul( iB(2,1:n),iC(1:n,1:n) )

! This would give 4, 4, 99
   ib(3,1) = 1
   iX2(1:n,1) = matmul( iB(2,1:n),iC(1:n,1:n) )

! Whereas, we should have 8, 8, 99
   if (any (iX1(1:n,1) .ne. (/8, 8, 99/))) call abort ()
   if (any (iX1 .ne. iX2)) call abort ()

! Make sure that the fix does not break transpose temporaries.
   iB = reshape((/(i, i = 1, 9)/),(/3,3/))
   ic = transpose (iB)
   iX1 = transpose (iB)
   iX1 = matmul (iX1, iC)
   iX2 = matmul (transpose (iB), iC)
   if (any (iX1 .ne. iX2)) call abort ()
   if (any (iX1 .ne. iChk)) call abort ()
end


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] gfortran: mathmul produces wrong result

[patchapp at dberlin dot org]

------- Comment #3 from patchapp at dberlin dot org  2006-06-14 17:00 -------
Subject: Bug number PR28005

A patch for this bug has been added to the patch tracker.
The mailing list url for the patch is
http://gcc.gnu.org/ml/gcc-patches/2006-06/msg00747.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] gfortran: mathmul produces wrong result

[pault at gcc dot gnu dot org]

------- Comment #4 from pault at gcc dot gnu dot org  2006-06-19 18:44 -------
I forgot to assign this to myself

Paul


-- 

pault at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |pault at gcc dot gnu dot org
                   |dot org                     |
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] gfortran: mathmul produces wrong result

[pault at gcc dot gnu dot org]

------- Comment #5 from pault at gcc dot gnu dot org  2006-06-20 04:31 -------
Subject: Bug 28005

Author: pault
Date: Tue Jun 20 04:30:48 2006
New Revision: 114802

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=114802
Log:
2006-06-20  Paul Thomas  <pault@gcc.gnu.org>

        PR fortran/25049
        PR fortran/25050
        * check.c (non_init_transformational): New function.
        (find_substring_ref): New function to signal use of disallowed
        transformational intrinsic in an initialization expression.
        (gfc_check_all_any): Call previous if initialization expr.
        (gfc_check_count): The same.
        (gfc_check_cshift): The same.
        (gfc_check_dot_product): The same.
        (gfc_check_eoshift): The same.
        (gfc_check_minloc_maxloc): The same.
        (gfc_check_minval_maxval): The same.
        (gfc_check_gfc_check_product_sum): The same.
        (gfc_check_pack): The same.
        (gfc_check_spread): The same.
        (gfc_check_transpose): The same.
        (gfc_check_unpack): The same.

        PR fortran/18769
        *intrinsic.c (add_functions): Add gfc_simplify_transfer.
        *intrinsic.h : Add prototype for gfc_simplify_transfer.
        *simplify.c (gfc_simplify_transfer) : New function to act as
        placeholder for eventual implementation.  Emit error for now.

        PR fortran/16206
        * expr.c (find_array_element): Eliminate condition on length of
        offset. Add bounds checking. Rearrange exit. Return try and
        put gfc_constructor result as an argument.
        (find_array_section): New function.
        (find_substring_ref): New function.
        (simplify_const_ref): Add calls to previous.
        (simplify_parameter_variable): Return on NULL expr.
        (gfc_simplify_expr): Only call gfc_expand_constructor for full
        arrays.

        PR fortran/20876
        * match.c (gfc_match_forall): Add missing locus to gfc_code.

2006-06-20  Paul Thomas  <pault@gcc.gnu.org>

        PR libfortran/28005
        * m4/matmul.m4: aystride = 1 does not uniquely detect the
        presence of a temporary transpose; an array element in the
        first dimension produces the same signature.  Detect this
        using the rank of a and add specific code.
        * generated/matmul_r4.c: Regenerate.
        * generated/matmul_r8.c: Regenerate.
        * generated/matmul_r10.c: Regenerate.
        * generated/matmul_r16.c: Regenerate.
        * generated/matmul_c4.c: Regenerate.
        * generated/matmul_c8.c: Regenerate.
        * generated/matmul_c10.c: Regenerate.
        * generated/matmul_c16.c: Regenerate.
        * generated/matmul_i4.c: Regenerate.
        * generated/matmul_i8.c: Regenerate.
        * generated/matmul_i16.c: Regenerate.

2006-06-20  Paul Thomas  <pault@gcc.gnu.org>

        PR fortran/16206
        * gfortran.dg/array_initializer_1.f90: New test.

        PR fortran/28005
        * gfortran.dg/matmul_3.f90: New test.


Added:
    trunk/gcc/testsuite/gfortran.dg/array_initializer_1.f90
    trunk/gcc/testsuite/gfortran.dg/matmul_3.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/check.c
    trunk/gcc/fortran/expr.c
    trunk/gcc/fortran/intrinsic.c
    trunk/gcc/fortran/intrinsic.h
    trunk/gcc/fortran/match.c
    trunk/gcc/fortran/simplify.c
    trunk/gcc/testsuite/ChangeLog
    trunk/libgfortran/ChangeLog
    trunk/libgfortran/generated/matmul_c10.c
    trunk/libgfortran/generated/matmul_c16.c
    trunk/libgfortran/generated/matmul_c4.c
    trunk/libgfortran/generated/matmul_c8.c
    trunk/libgfortran/generated/matmul_i16.c
    trunk/libgfortran/generated/matmul_i4.c
    trunk/libgfortran/generated/matmul_i8.c
    trunk/libgfortran/generated/matmul_r10.c
    trunk/libgfortran/generated/matmul_r16.c
    trunk/libgfortran/generated/matmul_r4.c
    trunk/libgfortran/generated/matmul_r8.c
    trunk/libgfortran/m4/matmul.m4


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] [4.1 only] gfortran: mathmul produces wrong result

[pault at gcc dot gnu dot org]

------- Comment #6 from pault at gcc dot gnu dot org  2006-06-24 10:55 -------
The patch did not apply to 4.1, so I will have to submit a back port.

Watch this space!

Paul


-- 

pault at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|gfortran: mathmul produces  |[4.1 only] gfortran: mathmul
                   |wrong result                |produces wrong result


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] [4.1 only] gfortran: matmul produces wrong result

[tobias dot burnus at physik dot fu-berlin dot de]

------- Comment #7 from tobias dot burnus at physik dot fu-berlin dot de  2006-07-05 09:23 -------
> The patch did not apply to 4.1, so I will have to submit a back port.
Thanks, Paul. I think this is the most serious bug in gfortran 4.1.x as it
silently produces wrong results.


-- 

tobias dot burnus at physik dot fu-berlin dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.1 only] gfortran: mathmul|[4.1 only] gfortran: matmul
                   |produces wrong result       |produces wrong result


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] [4.1 only] gfortran: matmul produces wrong result

[tobias dot burnus at physik dot fu-berlin dot de]

------- Comment #8 from tobias dot burnus at physik dot fu-berlin dot de  2006-08-24 19:43 -------
*ping*


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] [4.1 only] gfortran: matmul produces wrong result

[paulthomas2 at wanadoo dot fr]

------- Comment #9 from paulthomas2 at wanadoo dot fr  2006-08-24 21:40 -------
Subject: Re:  [4.1 only] gfortran: matmul produces wrong
 result

Tobias,

Seeing as it is you, I just put a first version on to cook; I simply 
copied everything below the line
if (GFC_DESCRIPTOR_RANK (retarray) == 1)
from 4.2 to 4.1.  I have to admit to not having kept up to speed on the 
interface differences between the two, which is why I have been so long 
in getting around to it.  In fact, I had half hoped that you would do 
it......!

Paul


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] [4.1 only] gfortran: matmul produces wrong result

[pault at gcc dot gnu dot org]

------- Comment #10 from pault at gcc dot gnu dot org  2006-08-29 04:57 -------
Subject: Bug 28005

Author: pault
Date: Tue Aug 29 04:57:29 2006
New Revision: 116553

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=116553
Log:
2006-08-29  Paul Thomas  <pault@gcc.gnu.org>

        PR fortran/28788
        REGRESSION FIX
        * symbol.c (gfc_use_derived): Never eliminate the symbol,
        following reassociation of use associated derived types.

2006-08-29  Paul Thomas  <pault@gcc.gnu.org>

        PR libfortran/28005
        * m4/matmul.m4: Working part of function ported from
        trunk.
        * generated/matmul_r4.c: Regenerate.
        * generated/matmul_r8.c: Regenerate.
        * generated/matmul_r10.c: Regenerate.
        * generated/matmul_r16.c: Regenerate.
        * generated/matmul_c4.c: Regenerate.
        * generated/matmul_c8.c: Regenerate.
        * generated/matmul_c10.c: Regenerate.
        * generated/matmul_c16.c: Regenerate.
        * generated/matmul_i4.c: Regenerate.
        * generated/matmul_i8.c: Regenerate.
        * generated/matmul_i16.c: Regenerate.

2006-08-29  Paul Thomas  <pault@gcc.gnu.org>

        PR fortran/28788
        * gfortran.dg/used_types_5.f90: New test.
        * gfortran.dg/used_types_6.f90: New test.

        PR fortran/28005
        * gfortran.dg/matmul_3.f90: New test.


Added:
    branches/gcc-4_1-branch/gcc/testsuite/gfortran.dg/matmul_3.f90
    branches/gcc-4_1-branch/gcc/testsuite/gfortran.dg/used_types_5.f90
    branches/gcc-4_1-branch/gcc/testsuite/gfortran.dg/used_types_6.f90
Modified:
    branches/gcc-4_1-branch/gcc/fortran/ChangeLog
    branches/gcc-4_1-branch/gcc/fortran/symbol.c
    branches/gcc-4_1-branch/libgfortran/ChangeLog
    branches/gcc-4_1-branch/libgfortran/generated/matmul_c10.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_c16.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_c4.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_c8.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_i16.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_i4.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_i8.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_r10.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_r16.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_r4.c
    branches/gcc-4_1-branch/libgfortran/generated/matmul_r8.c
    branches/gcc-4_1-branch/libgfortran/m4/matmul.m4


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005

[Bug fortran/28005] [4.1 only] gfortran: matmul produces wrong result

[pault at gcc dot gnu dot org]

------- Comment #11 from pault at gcc dot gnu dot org  2006-08-29 15:50 -------
I believe that does it.

Paul


-- 

pault at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28005