[Bug target/113431] [14] RISC-V rv64gcv vector: Runtime mismatch at -O3

["juzhe.zhong at rivai dot ai" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113431

--- Comment #8 from JuzheZhong <juzhe.zhong at rivai dot ai> ---
(In reply to Patrick O'Neill from comment #7)
> (In reply to JuzheZhong from comment #6)
> > (In reply to Patrick O'Neill from comment #5)
> > > (In reply to JuzheZhong from comment #4)
> > > > a[0][1] seems to be undefined value.
> > > 
> > > a is a global variable so the elements are initialized to 0. a[0][1] is
> > > within the bounds a[2][9].
> > 
> > Yes, I believe a[0][1] is 0. Then for all a[b][e] == a[0][1] == 0 should
> > always be
> > 1. 
> > 
> > Then if (!a[1][1]) should be false and then return 1;
> > 
> > I failed to see anything wrong here.
> 
> The value inside a[0][1] changes during the for loop. It starts out as 0 but
> is changed to 1 when b=0,e=1. The elements in a after b=0,e=1 should be 1.
> 
> Since a[1][1] is set after b=0,e=1 it should evaluate a[0][1]==0 to false
> and set a[1][1] to 0.

So I doubt it is not RISC-V backend issue. It's middle-end loop vectorizer
issue.
It's vectorized with evaluating whether it is 0 or not.

Let's wait for Andrew confirm it.