This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
RE: Shouldn't alias_sets_conflict_p be replaced with rtx_refs_may_alias_p?
For example, alias_sets_conflict_p will say the a[i]
is aliased with b[i]. It is both conservative and wrong.
void foo(int * restrict a, int * restrict b, int n)
{
int i;
for(i = 0; i < n; i++)
{
a[i] = b[i] * 100;
}
}
Bingfeng
> -----Original Message-----
> From: Steven Bosscher [mailto:stevenb.gcc@gmail.com]
> Sent: 04 August 2010 16:26
> To: Bingfeng Mei
> Cc: gcc@gcc.gnu.org; Richard Guenther
> Subject: Re: Shouldn't alias_sets_conflict_p be replaced with
> rtx_refs_may_alias_p?
>
> On Wed, Aug 4, 2010 at 2:02 PM, Bingfeng Mei <bmei@broadcom.com> wrote:
> > Hi,
> > alias_sets_conflict_p are still used in various places to determine
> > whether two memory accesses are aliased. It is based on unique set
> > number, which seems not correct with recent changes on alias
> > oracle. For example, in ddg.c cross-iteration memory dependence
> > is drawn by calling insn_alias_sets_conflict_p, which in turns calls
> > alias_sets_conflict_p and gets the wrong answer.
>
> Just to clarify: When you say "wrong", do you mean "incorrect" or "too
> conservative"?
>
> Ciao!
> Steven