[Patch, fortran] PR25378 - [Fortran 2003] maxloc for all-false mask

[Paul Thomas]

:ADDPATCH fortran:

This patch replaces the F95 processor dependent result for min/maxloc, 
in the event where the masks are all false, with the F2003 requirement 
that the result is zero.  Previously, gfortran returned 1.  Either is 
acceptable for F95 so it seems as well to implement Thomas Koenig's 
suggestion.  This is also consistent with commercial compilers.

The logic is:

set initial index = 0                             /* was 1 */
if (the max/min condition is met or index = 0)    /* rhs of or has been added */
  {
     set index to the current position
     update the value for comparison
  }


I have modified and extended the existing testcases and see no reason to 
add any more.

Regtested on FC3/Athlon.  OK for 4.1 and 4.2?

Paul

2006-03-09  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/25378
    * trans-intrinsic.c (gfc_conv_intrinsic_minmaxloc): Set the initial 
position to zero and
    modify the condition for updating it, to implement the F2003 
requirement for all(mask)
    is false.

2006-03-09  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/25378
    * libgfortran/m4/minloc1.m4: Set the initial position to zero and 
modify the condition for
    updating it, to implement the F2003 requirement for all(mask).eq.false.
    * libgfortran/m4/maxloc1.m4: The same.
    * libgfortran/m4/iforeach.m4: The same.
    * libgfortran/m4/minloc0.m4: The same.
    * libgfortran/m4/maxloc0.m4: The same.
    * libgfortran/generated/maxloc0_16_i16.c: Regenerated, together with 
41 others.
    * libgfortran/generated/minloc0_16_i16.c: Regenerated, together with 
41 others.

2006-03-09  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/25378
    * gfortran.fortran-torture/execute/intrinsic_mmloc_3.f90: Expand 
test to include more
    permuatations of mask and index.
    * testsuite/gfortran.dg/scalar_mask_1.f90: Modify last test to 
respond to F2003 spec.
    that the position returned for an all false mask && condition is zero.
  

PS I can list all 84 permutations of min/maxloc, if necessary but cannot 
see the value of it.


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: minmaxloc.diff
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20060308/f8396311/attachment.ksh>