Horton

Re: LLVM 16 (opaque pointers)

Thomas Munro <thomas.munro@gmail.com>

View thread
From: Thomas Munro <thomas.munro@gmail.com>
To: Andres Freund <andres@anarazel.de>
Cc: Ronan Dunklau <ronan.dunklau@aiven.io>, Devrim Gündüz <devrim@gunduz.org>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Fabien COELHO <coelho@cri.ensmp.fr>, Dmitry Dolgov <9erthalion6@gmail.com>
Date: 2023-10-20T23:02:51Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. jit: Adjust back-patch of f90b4a84 to 12 and 13.

  2. Log LLVM library version in configure output.

  3. jit: Changes for LLVM 17.

  4. jit: Supply LLVMGlobalGetValueType() for LLVM < 8.

  5. jit: Support opaque pointers in LLVM 16.

  6. jit: Reference function pointer types via llvmjit_types.c.

On Sat, Oct 21, 2023 at 11:12 AM Andres Freund <andres@anarazel.de> wrote:
> On 2023-10-21 10:48:47 +1300, Thomas Munro wrote:
> > On Thu, Oct 19, 2023 at 6:20 AM Thomas Munro <thomas.munro@gmail.com> wrote:
> > I see that Mark has also just enabled --with-llvm on some POWER Linux
> > animals, and they have failed in various ways.  The failures are
> > strangely lacking in detail.  It seems we didn't have coverage before,
> > and I recall that there were definitely versions of LLVM that *didn't*
> > work for our usage in the past, which I'll need to dredge out of the
> > archives.  I will try to get onto a cfarm POWER machine and see if I
> > can reproduce that, before and after these commits, and whose bug is
> > it etc.
>
> I'm quite sure that jiting did pass on ppc64 at some point.

Yeah, I remember debugging it on EDB's POWER machine.  First off, we
know that LLVM < 7 doesn't work for us on POWER, because:

https://www.postgresql.org/message-id/CAEepm%3D39F_B3Ou8S3OrUw%2BhJEUP3p%3DwCu0ug-TTW67qKN53g3w%40mail.gmail.com

That was fixed:

https://github.com/llvm/llvm-project/commit/a95b0df5eddbe7fa1e9f8fe0b1ff62427e1c0318

So I think that means that we'd first have to go through those animals
and figure out which ones have older LLVM, and ignore those results --
they just can't use --with-llvm.  Unfortunately there doesn't seem to
be any clue on the version from the paths used by OpenSUSE.  Mark, do
you know?

> > I doubt I can get anywhere near an s390x though, and we definitely had
> > pre-existing problems on that arch.
>
> IMO an LLVM bug, rather than a postgres bug, but I guess it's all a matter of
> perspective.
> https://github.com/llvm/llvm-project/issues/53009#issuecomment-1042748553

Ah, good to know about that.  But there are also reports of crashes in
released versions that manage to get passed that ABI-wobble business:

https://www.postgresql.org/message-id/flat/CAF1DzPXjpPxnsgySz2Zjm8d2dx7%3DJ070C%2BMQBFh%2B9NBNcBKCAg%40mail.gmail.com