[Bug tree-optimization/109088] GCC does not always vectorize conditional reduction

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109088

--- Comment #12 from JuzheZhong <juzhe.zhong at rivai dot ai> ---
(In reply to Richard Biener from comment #11)
> I don't think strip_nop_cond_scalar_reduction is the place to adjust here,
> maybe it's the caller.  I don't have time to dig into the specific issue
> right now but if we require scalar code adjustments then we need to perform
> those in if-conversion.
> 
> But to me it looks like allowing
> 
> > > STMT 1. tmp = a[i] + x;
> > > STMT 2. tmp2 = tmp + result_ssa_1;
> > > STMT 3. result_ssa_2 = mask ? tmp2 : result_ssa_1;
> 
> in vect_is_simple_reduction might also be a reasonable approach.  The
> use in the COND_EXPR isn't really a use - it's a conditional update.

Thanks Richi.

Enhancing vect_is_simple_reduction in loop vectorizer is also a good approach.
But I think it's better to recognize the scalar condition reduction
(if-conversion) as early as possible. Obviously, current if-conversion failed
to
recognize it as a feasible conditional reduction.

I think enhancing vect_is_simple_reduction is the approach that it's unlikely
we
can simplify the scalar code in if-converison to fit current loop vectorizer.

I believe we will eventually have to enhance both if-converison and loop
vectorizer in the future. And I prefer improving the if-conversion and working
on it. Will keep you posted.

Thanks a lot!