On 2/24/2026 3:54 AM, Richard Braun wrote:
Hello,
For a few months now, I've been working on the c6x target, and I've
produced a couple of patches, most of which straightforward to
understand, but a few of them, in particular those fixing issues in
the generic scheduler code, are more tricky, and I'd like feedback on
those to make sure they're right.
I've set up a GitHub branch [1] for convenience. The patches I'm most
concerned about are [2] and [3]. They should fix the ICE mentioned at
the end of PR 114910.
FYI, I'm building with --disable-libstdcxx-backtrace.
So I just tried the most straightforward LRA conversion. Essentially
drop the hook override and replace the reload_in_progress references it
lra_in_progress. That results in 645 testsuite regressions. Some are
ICEs in c6_expand_prologue, others are the more generic "excess errors"
kind of failure, which is likely an LRA failure. Point being, there's
more work that needs to be done to convert to LRA.
I've got no real interest in the c6x port other than as a test vehicle
for generic parts of the compiler. So I'm not going to chase these
down. But as an example:
c6x-sim: gcc: gcc.dg/tree-ssa/pr23848-1.c (internal compiler error: in
c6x_expand_prologue, at config/c6x/c6x.cc:2827)
If you want help keep this port alive, that's what I'd start with. I
doubt any of the maintainers are going to dive into your repository. We
deal with patches via this list, so you'd need to get the port converted
to LRA as one patchkit, post the patch here for review/approval. Once
that's working properly we can push it to the trunk, then you could work
through the rest of the changes you want to make with the same basic
process.
Jeff
