Hi,

This is the last patch needed for x32 support.
convert_memory_address_addr_space
is called to convert a memory address without overflow/underflow.  It
should be safe
to transform

(zero_extend:DI (plus:SI (FOO:SI) (const_int Y)))

to

(plus:DI (zero_extend:DI (FOO:SI)) (const_int Y))

GCC only works this way.  Any comments?

Thanks.

H.J.
----
On Sun, Aug 7, 2011 at 1:08 PM, H.J. Lu <hongjiu...@intel.com> wrote:
> Hi,
>
> We transform
>
> ptr_extend:DI (plus:SI (FOO:SI) (const_int Y)))
>
> to
>
> (plus:DI (ptr_extend:DI (FOO:SI)) (const_int Y))
>
> since this is how Pmode != ptr_mode is supported even if the resulting
> address may overflow/underflow.   It is also true for x32 which has
> zero_extend instead of ptr_extend.  I have tried different approaches
> to avoid transforming
>
> (zero_extend:DI (plus:SI (FOO:SI) (const_int Y)))
>
> to
>
> (plus:DI (zero_extend:DI (FOO:SI)) (const_int Y))
>
> without success.  This patch relaxes the condition to check
> POINTERS_EXTEND_UNSIGNED != 0 instead if POINTERS_EXTEND_UNSIGNED < 0
> to cover both ptr_extend and zero_extend. We can investigate a better
> approach for ptr_extend and zero_extend later.  For now, I believe it
> is the saftest way to support ptr_extend and zero_extend.
>
> Any comments?
>
> Thanks.
>
>
> H.J.
> ---
> gcc/
>
> 2011-08-06  H.J. Lu  <hongjiu...@intel.com>
>
>        PR middle-end/49721
>        * explow.c (convert_memory_address_addr_space): Also permute the
>        conversion and addition of constant for zero-extend.
>
> gcc/testsuite/
>
> 2011-08-06  H.J. Lu  <hongjiu...@intel.com>
>
>        PR middle-end/49721
>        * gfortran.dg/pr49721.f: New.
>
> diff --git a/gcc/explow.c b/gcc/explow.c
> index f8262db..ed2f621 100644
> --- a/gcc/explow.c
> +++ b/gcc/explow.c
> @@ -384,18 +384,23 @@ convert_memory_address_addr_space (enum machine_mode 
> to_mode ATTRIBUTE_UNUSED,
>
>     case PLUS:
>     case MULT:
> -      /* For addition we can safely permute the conversion and addition
> -        operation if one operand is a constant and converting the constant
> -        does not change it or if one operand is a constant and we are
> -        using a ptr_extend instruction  (POINTERS_EXTEND_UNSIGNED < 0).
> +      /* FIXME: For addition, we used to permute the conversion and
> +       * addition operation only if one operand is a constant and
> +        converting the constant does not change it or if one operand
> +        is a constant and we are using a ptr_extend instruction
> +        (POINTERS_EXTEND_UNSIGNED < 0) even if the resulting address
> +        may overflow/underflow.  We relax the condition to include
> +        zero-extend (POINTERS_EXTEND_UNSIGNED > 0) since the other
> +        parts of the compiler depend on it.  See PR 49721.
> +
>         We can always safely permute them if we are making the address
>         narrower.  */
>       if (GET_MODE_SIZE (to_mode) < GET_MODE_SIZE (from_mode)
>          || (GET_CODE (x) == PLUS
>              && CONST_INT_P (XEXP (x, 1))
> -             && (XEXP (x, 1) == convert_memory_address_addr_space
> -                                  (to_mode, XEXP (x, 1), as)
> -                 || POINTERS_EXTEND_UNSIGNED < 0)))
> +             && (POINTERS_EXTEND_UNSIGNED != 0
> +                 || XEXP (x, 1) == convert_memory_address_addr_space
> +                                       (to_mode, XEXP (x, 1), as))))
>        return gen_rtx_fmt_ee (GET_CODE (x), to_mode,
>                               convert_memory_address_addr_space
>                                 (to_mode, XEXP (x, 0), as),
> diff --git a/gcc/testsuite/gfortran.dg/pr49721.f 
> b/gcc/testsuite/gfortran.dg/pr49721.f
> new file mode 100644
> index 0000000..39e2ed7
> --- /dev/null
> +++ b/gcc/testsuite/gfortran.dg/pr49721.f
> @@ -0,0 +1,35 @@
> +! PR middle-end/49721
> +! { dg-do compile }
> +! { dg-options "-O3 -funroll-loops" }
> +
> +      subroutine midbloc6(c,a2,a2i,q)
> +      parameter (ndim2=6)
> +      parameter (ndim=3)
> +      dimension ri(ndim2),cr(ndim2,ndim2),xj(ndim2,ndim2),q(*)
> +     @,sai(ndim2,ndim2),cm(ndim2,ndim2),w(ndim2,ndim2)
> +      dimension vr(ndim2,ndim2),vi(ndim2,ndim2),s1(ndim2,ndim2),p(ndim)
> +      dimension xq(6),qb(2),qc(2),ifl(6),iplane(3)
> +      save
> +      call eig66(cr,rr,ri,vr,vi)
> +      xq(i)=asin(ri(i))/x2pi
> +      i9=6
> +      qb(1)=q(1)/x2pi
> +        do 180 i=1,2
> +          do 170 j=1,6
> +  120       if(xq(j)) 130,190,140
> +  130       if(qb(i)-0.5d0) 160,150,150
> +  140       if(qb(i)-0.5d0) 150,150,160
> +  150       continue
> +            tst=abs(abs(qb(i))-abs(xq(j)))
> +  160       continue
> +  170     continue
> +          iplane(i)=k
> +  180   continue
> +  190   continue
> +      n1=iplane(3)
> +      if(i9.eq.6) then
> +        z=vr(1,n1)*vi(2,n1)-vr(2,n1)*vi(1,n1)+vr(3,n1)*vi(4,n1)-vr(4,n1)
> +      endif
> +      sai(6,i)=vi(i,n1)/z
> +      call dacond6(a2,zero)
> +      end
>



-- 
H.J.

Reply via email to