https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123108
Jeffrey A. Law <law at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target Milestone|16.0 |---
--- Comment #4 from Jeffrey A. Law <law at gcc dot gnu.org> ---
So we discussed this bug in the patchwork call last week.
We don't see a great way forward right now. The core problem is the allocators
can do limited substitutions of equivalences without checking the insn matches.
Normally we'd have a constraint that would then reject the constant and force
reloading. But the patterns in autovec.md don't have constraints.
Either we need to fix IRA/LRA to avoid these substitutions or we need to add
constraints to all those patterns. In both cases it feels like the risk/reward
isn't great for gcc-16 inclusion. So current expectation is this will not get
fixed in gcc-16.
We don't have a target milestone for gcc-17 yet, so I've just reset that value
for now.