This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [autovect] [patch] Dependence tests and dimensions
- From: Daniel Berlin <dberlin at dberlin dot org>
- To: Ira Rosen <IRAR at il dot ibm dot com>
- Cc: gcc-patches at gcc dot gnu dot org, Sebastian Pop <sebastian dot pop at cri dot ensmp dot fr>
- Date: Sun, 19 Jun 2005 08:40:24 -0400
- Subject: Re: [autovect] [patch] Dependence tests and dimensions
- References: <OFB2206743.39067934-ONC2257022.00429226-C2257025.00336F31@il.ibm.com>
On Sun, 2005-06-19 at 12:21 +0300, Ira Rosen wrote:
>
>
>
> This patch removes a dependence test based on DRs dimensions: while being
> correct for ARRAY_REFs, it is incorrect to decide that two data-refs are
> independent if their dimensions numbers differ for the case when both
> ARRAY_REFs and INDIRECT_REFs are tested.
> E.g.,
> for i
> for j
> *((int *)p + i + j) = a[i][j];
>
> The access function of the pointer access is {j_init, + 4B}_2 relative to
> 'p + i', and the access functions of the array are '{i_init, + i_step}_1
> and {j_init, +, j_step}_2'. Therefore, 'a[i][j] is two-dimensional and 'p'
> is one-dimensional. But, of course, 'p' can point to 'a', and we need to
> apply alias analysis to check this.
Here, p can't point to a, unless you've written the example wrong.
a is the equivalent of int **
p is the equivalent of int *
For p to have taken the address of a and be properly dereferenced, it
would have to be int ***.