[Bug target/115258] [14/15 Regression] register swaps for vector perm in some cases after r14-6290

[rsandifo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115258

Richard Sandiford <rsandifo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rsandifo at gcc dot gnu.org

--- Comment #2 from Richard Sandiford <rsandifo at gcc dot gnu.org> ---
I agree with Andrew that this seems mostly to be luck.  On that basis, I'm not
sure (either way) whether we should backport the fix.

Now that we're hopefully getting better at tracking and allocating subregs, it
probably makes sense to allow the split before reload.  Doing that (and
adjusting the split accordingly) seems to fix the testcase.