[Bug middle-end/106081] missed vectorization

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106081

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
                 CC|                            |rguenth at gcc dot gnu.org
             Blocks|                            |53947

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Might be a bit easier case than PR103999 since the accumulation is in 'double'
here.  In fact it should already work, but we fail to SLP the reduction
because

t.c:14:24: missed:   Build SLP failed: not grouped load _1 = *k_50;

and non-SLP as well:

t.c:14:24: note:   ==> examining statement: _1 = *k_50;
t.c:14:24: missed:   multiple types with negative step.
t.c:14:24: missed:   not falling back to elementwise accesses
t.c:16:30: missed:   not vectorized: relevant stmt not supported: _1 = *k_50;
t.c:14:24: missed:  bad operation or unsupported loop bound.

changing k-- to k++ lets us vectorize the code but in an awkward way
(SLP still fails on us so we get interleaving).

So the first thing to fix is SLP of non-grouped loads where I think we also
have a bug for.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53947
[Bug 53947] [meta-bug] vectorizer missed-optimizations