Re: [PATCH][RFC] Add PAREN_EXPR, make flag_associative_math the default for Fortran

[dominiq at lps dot ens dot fr (Dominique Dhumieres)]

Richard,

Happy New Year and thanks for the patch.
In my opinion gfortran shall obey the parentheses whatever
the options are. I have applied the patch on Intel Darwin9
and I have a few glitches:

(1) some ICEs on tests such as:

  character(len=1) :: string
  print *, transfer(((transfer(string,"x",1))), "x")
  end

or

! { dg-do compile }
! PR fortran/29403
program p
   character(len=10) a, b, c
   integer i
   i = 1
   print ('(I0)'), i
   a = '(I0,'
   b = 'I2,'
   c = 'I4)'
   call prn(a, b, c, i)
!   print (1,*), i       ! { dg-error "in PRINT statement" }
end program p

subroutine prn(a, b, c, i)
  integer i
  character(len=*) a, b, c
  print (a//(b//c)), i, i, i
  print trim(a//trim(b//c)), i, i, i
end subroutine prn

pr31608_1.f90: In function 'MAIN__':
pr31608_1.f90:1: internal compiler error: in gfc_conv_unary_op, at fortran/trans-expr.c:592

or

print_1_db.f90: In function 'prn':
print_1_db.f90:15: internal compiler error: in gfc_conv_unary_op, at fortran/trans-expr.c:592

(2) "stuff like (x + 1e52) - 1e52" seems to work, but my checking test does not work
as expected and I'll have to investigate in more details.

(3) I have tried to compile the file coming from:

C***********************************************************************
C                TEST SUITE FOR VECTORIZING COMPILERS                  *
C                        (File 2 of 2)                                 *
C                                                                      *
C  Version:   2.0                                                      *
C  Date:      3/14/88                                                  *
C  Authors:   Original loops from a variety of                         *
C             sources. Collection and synthesis by                     *
C                                                                      *
C             David Callahan  -  Tera Computer                         *
C             Jack Dongarra   -  University of Tennessee               *
C             David Levine    -  Argonne National Laboratory           *
C***********************************************************************
C  Version:   3.0                                                      *
C  Date:      1/4/91                                                   *
C  Authors:   David Levine    -  Executable version                    *
C***********************************************************************
...

and the patched gfortran gives an ICE for three subroutines and miss
a few vectorization. One subroutine giving an ICE at -O2 and above is

      subroutine s125(ntimes,ld,n,ctime,dtime,a,b,c,d,e,aa,bb,cc)
C
C     induction variable recognition
C     induction variable in two loops; collapsing possible
C
      integer ntimes,ld,n,i,nl,j,k,nn
      real a(n),b(n),c(n),d(n),e(n),aa(ld,n),bb(ld,n),cc(ld,n)
      real t1,t2,chksum,ctime,dtime,cs1d,array
      parameter(nn= 1000)
      common/cdata /array(nn*nn)
          do j= 1,n
                array((j-1)*n+1:j*n)= aa(:n,j)+bb(:n,j)*cc(:n,j)
            enddo
      return
      end

Thanks again for the work

Dominique

PS I think the file I am refering to in (3) comes from

http://softwarecommunity.intel.com/Wiki/HighPerformanceComputing/688.htm