https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126640
Bug ID: 126640
Summary: riscv: -mtune can introduce misaligned vectors.
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: rdapp at gcc dot gnu.org
CC: kito at gcc dot gnu.org, law at gcc dot gnu.org
Target Milestone: ---
Target: riscv
While looking into the the Oilsm proposal, I noticed that with e.g.
-mtune=generic-ooo (which enables misaligned vectors) we do emit misaligned
vectors, regardless of the -march. IMHO that's wrong and I know that I
introduced this :) -march=foo -mtune=bar code must always be executable on any
foo architecture even if it's slow. Potentially faulting vector accesses are
contrary to this principle. I don't think the blast radius of correcting it
would be too large. Perhaps we should just define a "generic-ooo" CPU/arch in
addition.
One way or another, right now -mtune can create executables that won't run on
CPUs that trap on misaligned vectors.
Kito, how is the situation on the LLVM side here?
