[3.8 Release] Release status

According to the schedule (e.g. on the right on llvm.org), we should
have tagged the release by now, but we haven't, so we're officially
behind schedule. I'm still optimistic that we can wrap this up pretty
soon, though.

This is what's blocking us:

- PR26509: Crash in InnerLoopVectorizer::vectorizeLoop()
  I'm waiting to hear what Cong comes up with, otherwise we can revert
r255691 on the branch

- Shrink-wrapping vs TLS: Davide and Quentin are working on it

- PR26600: Loop vectorization creates an unsafe out-of-bounds load
  There's a patch out: http://reviews.llvm.org/D17332
  But no comments yet.. Hal?

- PR26564: Performance regression in AA
  Patch in review, but it makes me a little uneasy since it's big and
requires pulling in some refactoring patches too :-/

- ARM: fix VFP asm constraints: http://reviews.llvm.org/D17349
  It's not a regression, but I'll take it if it lands real soon.

- PR26500: shrink-wrapping vs PPC
  Patch in review: http://reviews.llvm.org/D17294
  Looks like it's moving along.

- PR26081: Assertion failed: (BitWidth == RHS.BitWidth && ...
  I believe Matthias is working on it?
  Will revert r252839 to unblock otherwise.

- PR26485: regression lowering TLS access in C on Darwin
  Is no one looking? :frowning:

If you're on one of these bugs or code reviews, etc., please try to
prioritize them if you can.

Also, please let me know if my list is missing something.

Cheers,
Hans

http://reviews.llvm.org/D17427

I'd suggest that https://llvm.org/bugs/show_bug.cgi?id=26554 should be
on that list.

- John Kåre

Hello,

this is probably not a blocker, but a rather small patch which is on review for quite some time:

http://reviews.llvm.org/D13811

Just a suggestion integrating it into the release.

Best Regards,
Florian

Hi Hans,

I think PR26369 should be on the list but we could potentially make do without it (see below). Without it we get 23 failures on libcxx tests caused by failing to link the tests with libatomic but libcxx itself is fine.

One thing that's unclear is whether these failures are really regressions or not. My logs for 3.7.1 don't show any failures which would suggest they are regressions. However I've just run 'make check-libcxx' on that old build and I get the same failures I see on 3.8. I'm currently digging into the 3.7.1 build to see if I can explain why the log differs from what I currently get.

- PR26600: Loop vectorization creates an unsafe out-of-bounds load
  There's a patch out: http://reviews.llvm.org/D17332
  But no comments yet.. Hal?

This looks like a serious bug but Hal's idea is still unproven. I
think we should land the patch as it is, since it disables the bad
behaviour, and think about a fix later. Depending on how complex the
fix is, we might not even merge it into 3.8.x later on, but we need
the fix in 3.8.0.

- ARM: fix VFP asm constraints: http://reviews.llvm.org/D17349
  It's not a regression, but I'll take it if it lands real soon.

No brainer. Accepted on trunk by compnerd, I'll merge into release_38 shortly.

cheers,
--renato

From: "Renato Golin" <renato.golin@linaro.org>
To: "Hans Wennborg" <hans@chromium.org>
Cc: "llvm-dev" <llvm-dev@lists.llvm.org>, "cfe-dev" <cfe-dev@lists.llvm.org>, "LLDB Dev" <lldb-dev@lists.llvm.org>,
"openmp-dev (openmp-dev@lists.llvm.org)" <openmp-dev@lists.llvm.org>, release-testers@lists.llvm.org, "Cong Hou"
<congh@google.com>, "Davide Italiano" <davide@freebsd.org>, "Quentin Colombet" <qcolombet@apple.com>, "Hal Finkel"
<hfinkel@anl.gov>, "Chandler Carruth" <chandlerc@google.com>, "JF Bastien" <jfb@google.com>, "Nemanja Ivanovic"
<nemanja.i.ibm@gmail.com>, "Matthias Braun" <matze@braunis.de>
Sent: Friday, February 19, 2016 9:33:16 AM
Subject: Re: [lldb-dev] [3.8 Release] Release status

> - PR26600: Loop vectorization creates an unsafe out-of-bounds load
> There's a patch out: http://reviews.llvm.org/D17332
> But no comments yet.. Hal?

This looks like a serious bug but Hal's idea is still unproven.

I feel your statement misrepresents my review of the patch. What I said in the review was:

[from review]
How complicated would it be, instead of bailing out when we have a group without the last member, to peel off the last vector iteration instead (i.e. jump to the scalar tail loop one vector-loop iteration "early")?

It seems like that would be a better solution (although, if you agree, but it seems too complicated to implement for the release branch, I'm fine with taking this (and pulling it into the release branch), and then implementing the better solution in trunk only).
[end from review]

What we have now is one easy fix, and I simply asked if we could get a much better easy fix in short order (i.e. for 3.8.0). Regardless, I feel I was quite clear in the review that this fix is otherwise fine, and we definitely need something for 3.8.0. It is a serious bug.

Thanks again,
Hal

I've changed the lit arguments to '-v' on my old 3.7.1 build and re-run check-all. I can confirm that these failures aren't regressions since LLVM 3.7.1 did not actually run any libc++ tests. Looking through the generated makefiles it seems that check-libcxx wasn't part of check-all.

- PR26600: Loop vectorization creates an unsafe out-of-bounds load
  There's a patch out: http://reviews.llvm.org/D17332
  But no comments yet.. Hal?

This looks like a serious bug but Hal's idea is still unproven. I
think we should land the patch as it is, since it disables the bad
behaviour, and think about a fix later. Depending on how complex the
fix is, we might not even merge it into 3.8.x later on, but we need
the fix in 3.8.0.

r261341.

- ARM: fix VFP asm constraints: http://reviews.llvm.org/D17349
  It's not a regression, but I'll take it if it lands real soon.

No brainer. Accepted on trunk by compnerd, I'll merge into release_38 shortly.

r261343.

cheers,
--renato

Hi Hans,

Thanks for pushing this forward.

According to the schedule (e.g. on the right on llvm.org), we should
have tagged the release by now, but we haven't, so we're officially
behind schedule. I'm still optimistic that we can wrap this up pretty
soon, though.

This is what's blocking us:

- PR26509: Crash in InnerLoopVectorizer::vectorizeLoop()
I'm waiting to hear what Cong comes up with, otherwise we can revert
r255691 on the branch

- Shrink-wrapping vs TLS: Davide and Quentin are working on it

http://reviews.llvm.org/D17427

- PR26600: Loop vectorization creates an unsafe out-of-bounds load
There's a patch out: http://reviews.llvm.org/D17332
But no comments yet.. Hal?

- PR26564: Performance regression in AA
Patch in review, but it makes me a little uneasy since it's big and
requires pulling in some refactoring patches too :-/

- ARM: fix VFP asm constraints: http://reviews.llvm.org/D17349
It's not a regression, but I'll take it if it lands real soon.

- PR26500: shrink-wrapping vs PPC
Patch in review: http://reviews.llvm.org/D17294
Looks like it's moving along.

- PR26081: Assertion failed: (BitWidth == RHS.BitWidth && ...
I believe Matthias is working on it?
Will revert r252839 to unblock otherwise.

- PR26485: regression lowering TLS access in C on Darwin
Is no one looking? :frowning:

I’ll have a look.

Cheers,
-Quentin

Hi Hans,

Hi Hans,

Thanks for pushing this forward.

According to the schedule (e.g. on the right on llvm.org), we should
have tagged the release by now, but we haven’t, so we’re officially
behind schedule. I’m still optimistic that we can wrap this up pretty
soon, though.

This is what’s blocking us:

  • PR26509: Crash in InnerLoopVectorizer::vectorizeLoop()
    I’m waiting to hear what Cong comes up with, otherwise we can revert
    r255691 on the branch

  • Shrink-wrapping vs TLS: Davide and Quentin are working on it

http://reviews.llvm.org/D17427

  • PR26600: Loop vectorization creates an unsafe out-of-bounds load
    There’s a patch out: http://reviews.llvm.org/D17332
    But no comments yet… Hal?

  • PR26564: Performance regression in AA
    Patch in review, but it makes me a little uneasy since it’s big and
    requires pulling in some refactoring patches too :-/

  • ARM: fix VFP asm constraints: http://reviews.llvm.org/D17349
    It’s not a regression, but I’ll take it if it lands real soon.

  • PR26500: shrink-wrapping vs PPC
    Patch in review: http://reviews.llvm.org/D17294
    Looks like it’s moving along.

  • PR26081: Assertion failed: (BitWidth == RHS.BitWidth && …
    I believe Matthias is working on it?
    Will revert r252839 to unblock otherwise.

  • PR26485: regression lowering TLS access in C on Darwin
    Is no one looking? :frowning:

Should be fixed as revision 261384 on trunk.

I leave the merge to you.

Thanks,
-Quentin

Many thanks to everyone who helped out after this email yesterday!

Here's an update on the remaining issues. As you can see, it's a much
shorter list :slight_smile: I'm hoping to tag rc3 very soon.

Thanks again,
Hans

- Shrink-wrapping vs TLS: Davide and Quentin are working on it

Fix is in r261387, will merge once it's baked in the tree for a bit.

- PR26564: Performance regression in AA
  Patch in review, but it makes me a little uneasy since it's big and
requires pulling in some refactoring patches too :-/

Keeping an eye on it, but it's not blocking.

- PR26500: shrink-wrapping vs PPC
  Patch in review: http://reviews.llvm.org/D17294

Looks like it's pretty much ready to land.

- PR26485: regression lowering TLS access in C on Darwin

Fix is in r261384, will merge after it's been in-tree for a while.

- r261297 - Implement the likely resolution of core issue 253.
  New from yesterday. Might want to merge this. Post-commit review
  is still ongoing.

Hi Hans,

I recently discovered this regression in Clang 3.8, coming from 3.7.1:

https://llvm.org/bugs/show_bug.cgi?id=26398
Regression: -Wnull-dereference (enabled by default) even applies to
expressions that are not evaluated at run-time.

If we're not going to fix this warning in time for 3.8.0-release,
should we at least consider disabling -Wnull-dereference by default?

Thanks,

I see that Aaron has a patch out at http://reviews.llvm.org/D17507
Let's see how that goes.

Thanks,
Hans

I had hoped to tag rc3 today (I feel like I've said this a lot
lately), but it's at least really, really close. I'm waiting for:

- r261297 - Implement the likely resolution of core issue 253.
  Still in post-commit review.

- D17507 - The controlling expression for _Generic is unevaluated
  New for today. Waiting for review.

Thanks,
Hans

I originally wanted this one but the fix involved even more discussion
so I guess it would be safer to put it in 3.8.1 instead.

Thanks,
ismail

This one should hopefully be relatively straight-forward to review.

~Aaron

Right. Since this seems complicated, and not strictly fixing a
regression, I'm not going to wait for it.

Thanks,
Hans