PR33749 - Wrong evaluation of expressions in lhs of assignment statements

[Paul Richard Thomas]

This is by way of an appeal...

This bug only appears in -m32 or, rather, it disappears in -m64.  What
is odd is that an entire block of code produced by the scalarizer
appears and disappears according to the mode.  What is even odder is
that I cannot even figure out how the correct code is ever produced.

This:

  integer(4) :: p(4) = (/2,4,1,3/)
  p(p) = (/(i, i = 1, 4)/)
  print *, p
end

with -m64 produces:
  static int4 p[4] = {2, 4, 1, 3};
  int4 i;
  static int4 options.0[7] = {68, 127, 0, 0, 0, 1, 0};

  _gfortran_set_options (7, (void *) &options.0);
  {
    int8 A.3[4];
    struct array1_int8 atmp.2;
    static int4 A.1[4] = {1, 2, 3, 4};

    atmp.2.dtype = 521;
    atmp.2.dim[0].stride = 1;
    atmp.2.dim[0].lbound = 0;
    atmp.2.dim[0].ubound = 3;
    atmp.2.data = (void *) &A.3;
    atmp.2.offset = 0;
    {
      int8 S.4;

      S.4 = 0;
      while (1)
        {
          if (S.4 > 3) goto L.1;
          (*(int8[0:] *) atmp.2.data)[S.4] = (int8) p[S.4];
          S.4 = S.4 + 1;
        }
      L.1:;
    }
    {
      int8 S.5;

      S.5 = 0;
      while (1)
        {
          if (S.5 > 3) goto L.2;
          {
            int8 D.875;

            D.875 = (*(int8[0:] *) atmp.2.data)[S.5];
            p[D.875 + -1] = A.1[S.5];
          }
          S.5 = S.5 + 1;
        }
      L.2:;
    }
  }
........+ the io.....

but the block that produces D.875 goes away in -m32 mode, so that the
assignment becomes, incorrectly:

            D.513 = (*(int4[0:] *) parm.2.data)[parm.2.dim[0].stride *
NON_LVALUE_EXPR <S.3>];
            p[D.513 + -1] = A.1[S.3];

an anybody cast any light on this?  It's driving me bonkers...

Paul



-- 
The knack of flying is learning how to throw yourself at the ground and miss.
       --Hitchhikers Guide to the Galaxy