Re: [PATCH] Fix SLP vectorization of shifts (PR tree-optimization/48616)

[Ira Rosen <IRAR at il dot ibm dot com>]

Jakub Jelinek <jakub@redhat.com> wrote on 17/04/2011 05:26:14 PM:
>
> On Sun, Apr 17, 2011 at 11:30:31AM +0300, Ira Rosen wrote:
> > We already have this check in vect_build_slp_tree(). It didn't work for
the
> > testcase because it doesn't take into account the type of definition. I
> > agree that it's better to move it here and base the vector/scalar
decision
> > on it, but please remove the now redundant check from
vect_build_slp_tree
> > ().
>
> I've tried to add this to my patch, unfortunately slp-36.c test then
fails
> on both x86_64-linux and i686-linux.
>
> The problem is that during the analysis vectorizable_shift is called with
> NULL slp_node, but later on it is called with non-NULL:

Right, sorry, I haven't thought about this. The vect_build_slp_tree check
is done during the analysis and it checks that if vector shift argument is
not supported, the arguments should be equal. And the vectorizable_shift
SLP check is done only during transformation. So, your original patch is
OK.

Thanks,
Ira