[Bug rtl-optimization/92796] [10 Regression] ICE in lra_assign, at lra-assigns.c:1646 on powerpc64le-linux-gnu

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92796

--- Comment #10 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
(In reply to Vladimir Makarov from comment #7)

>
> 
> I'm guessing this was never a problem before I added the code to not add
> conflicts for copies.  Before then, any two pseudos/registers that were live
> simultaneously were made to conflict, which would have hid this bug.

I believe this bug existed before.  For example, the initial assignment (and
some LRA transformations) does not guarantee correctness of already made
assignments.

Therefore there is flag lra_risky_transformations_p (I am going to change this
name) to deal with such situations.  Simply the current situation was not
considered before.

I have a patch and after some testing I'll commit it (probably tomorrow).