[Bug tree-optimization/92772] wrong code vectorizing masked max

[ams at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92772

Andrew Stubbs <ams at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P5
           Severity|critical                    |minor

--- Comment #7 from Andrew Stubbs <ams at gcc dot gnu.org> ---
Thank you for the help, Richard and Richard.

I'm electing not to fix this bug, but instead have worked around it for amdgcn
by implementing the fold_extract_last pattern.

There's two main reasons for this:

1. I'm not confident I can see how to fix it properly, so it's going to take me
a long time to get it right.

2. The code that it generates is horrible (and the fix will only make it
worse), so I would want to implement fold_extract_last anyway. This would leave
this code unused by amdgcn, and the time spent fixing it wasted effort.

I've therefore downgraded the priority and severity on this bug, and committed
a pointer into the code to assist anybody tripping over the same issue in
future.

https://gcc.gnu.org/ml/gcc-patches/2019-12/msg01191.html

The bug still exists, but is now no longer reproducible on amdgcn.