[7.0.0 Release] rc2 has been tagged

Project Infrastructure Release Testers
1

Dear testers,

7.0.0-rc2 was just tagged (from branch revision r340437).

There have been a bunch of merges since rc1, and hopefully many of the
issues with the previous candidate are fixed in this one.

Please run the test script, share the results, and upload binaries.

I will publish source tarballs, docs, and binaries on the web page
once they're ready.

Thanks,
Hans

2

Note: to upload, we now need to use prereleases-origin.llvm.org
instead of prereleases.llvm.org. This is because the latter now points
to the CDN.

3

Hi Hans,

I’ve uploaded Darwin/x86_64 binaries to prereleases-origin.llvm.org.

$ shasum -a 256 clang+llvm-7.0.0-rc2-x86_64-apple-darwin.tar.xz
88d18baf8a066395c6d2a54a731aff56f3cb015712071b73f4b0bcd2347eaa76 clang+llvm-7.0.0-rc2-x86_64-apple-darwin.tar.xz

There were no new issues, but llvm.org/PR38473 is still present. I reproduced one of the crashes locally and attached a backtrace to the bug report.

thanks,
vedant

4

Windows:

$ sha1sum LLVM-7.0.0-rc2*.exe
e9e241eea5ded7692b0d57a4943b56dce293634f LLVM-7.0.0-rc2-win32.exe
75ebb78fa75254287090baa4ab0b0911fc16fa47 LLVM-7.0.0-rc2-win64.exe

They were built with the attached batch file.

build_llvm_700-rc2.bat|attachment (4.09 KB)

5

Don’t forget to compile (everything if possible, but especially LLVM) as a shared library on Windows.

the installed toolchain is about 10 times bigger than the executable, because the static library is being compressed, which means on disk it’s taking up far more space than it needs.

6

Also, can you include Polly in the build as well?

7

FYI, I have uploaded CentOS6 RPMs.
https://github.com/llvm-project/llvm-project-20170507/releases/tag/RELEASE_700%2Frc2

8

By reverting r323281 locally, on top of rc2, I was now able to build and test for i386-freebsd11 too.

Main test results on amd64-freebsd11 look better, roughly 2000 less failures, mostly due to the libc++ get_timespec fix:

  Expected Passes : 52409 (rc1: 50388)
  Expected Failures : 232 (rc1: 233)
  Unsupported Tests : 3687 (rc1: 3687)
  Unexpected Passes : 1 (rc1: 1)
  Unexpected Failures: 491 (rc1: 2490)

Test-suite test results on amd64-freebsd11:

  Expected Passes : 845
  Unexpected Failures: 61

Test results on i386-freebsd11:

  Expected Passes : 50186
  Expected Failures : 226
  Unsupported Tests : 2502
  Unexpected Failures: 306

Unfortunately the test-suite doesn't build on i386, since quite a few of its components requires SSE2, and the build shows many errors like:

/home/dim/llvm/7.0.0/rc2/test-suite.src/Bitcode/Benchmarks/Halide/blur/driver.cpp:37:29: error: always_inline function '_mm_set1_epi16' requires target feature 'sse2', but would be inlined into function 'blur_fast' that is compiled without support for 'sse2'
        __m128i one_third = _mm_set1_epi16(21846);
                            ^

Uploaded:

SHA256 (clang+llvm-7.0.0-rc2-amd64-unknown-freebsd11.tar.xz) = 67cddaea2123bd7674c5f1ab8092cd7fb2b43ab03dd235469d117782595128fa
SHA256 (clang+llvm-7.0.0-rc2-i386-unknown-freebsd11.tar.xz) = cfaadd88255fc8f6fd3d425f61d8c8ef67c8776a0d5a4d6acd6da8cef95085bb

-Dimitry

9

If we had a way of building e.g. at least LLVM as a dll, maybe we
could do this, but currently the binaries are all statically linked
and there is no plan to change this.

10

I don't know how well Polly works on Windows. We've never had it in
the monthly snapshots (https://www.llvm.org/builds/), and that would
be a pre-requisite before bringing it into releases.

Tobias: does Polly work on Windows, and should we try to include it in
the snapshots?

11

7.0.0-rc2 was just tagged (from branch revision r340437).

There have been a bunch of merges since rc1, and hopefully many of the
issues with the previous candidate are fixed in this one.

By reverting r323281 locally, on top of rc2, I was now able to build and test for i386-freebsd11 too.

Thanks! The revert will be part of rc3.

Main test results on amd64-freebsd11 look better, roughly 2000 less failures, mostly due to the libc++ get_timespec fix:

  Expected Passes : 52409 (rc1: 50388)
  Expected Failures : 232 (rc1: 233)
  Unsupported Tests : 3687 (rc1: 3687)
  Unexpected Passes : 1 (rc1: 1)
  Unexpected Failures: 491 (rc1: 2490)

Test-suite test results on amd64-freebsd11:

  Expected Passes : 845
  Unexpected Failures: 61

Test results on i386-freebsd11:

  Expected Passes : 50186
  Expected Failures : 226
  Unsupported Tests : 2502
  Unexpected Failures: 306

Unfortunately the test-suite doesn't build on i386, since quite a few of its components requires SSE2, and the build shows many errors like:

/home/dim/llvm/7.0.0/rc2/test-suite.src/Bitcode/Benchmarks/Halide/blur/driver.cpp:37:29: error: always_inline function '_mm_set1_epi16' requires target feature 'sse2', but would be inlined into function 'blur_fast' that is compiled without support for 'sse2'
        __m128i one_third = _mm_set1_epi16(21846);
                            ^

The same problem must have been present also with 6.0.0 right?

12

...

Unfortunately the test-suite doesn't build on i386, since quite a few of its components requires SSE2, and the build shows many errors like:

/home/dim/llvm/7.0.0/rc2/test-suite.src/Bitcode/Benchmarks/Halide/blur/driver.cpp:37:29: error: always_inline function '_mm_set1_epi16' requires target feature 'sse2', but would be inlined into function 'blur_fast' that is compiled without support for 'sse2'
       __m128i one_third = _mm_set1_epi16(21846);
                           ^

The same problem must have been present also with 6.0.0 right?

Yes, this has apparently always been the case. I'll make a note to either fix it post release, or to selectively skip building those tests for i386.

-Dimitry

13

I’m just curious why you don’t plan on using dlls? Xcode ships with libclang as a dylib, i wouldnt think windows would be a big leap?

14

That libclang dylib just implements an interface used for source code
indexing and such (https://clang.llvm.org/doxygen/group__CINDEX.html).
It's not used by the clang binary.

15

Hello,

Dear testers,

7.0.0-rc2 was just tagged (from branch revision r340437).

There have been a bunch of merges since rc1, and hopefully many of the
issues with the previous candidate are fixed in this one.

Please run the test script, share the results, and upload binaries.

I will publish source tarballs, docs, and binaries on the web page
once they're ready.

Sorry for the latency.
It took sometime for the ftpmasters in Debian to accept the package and to adjust
the CI for https://apt.llvm.org/.
Now, all Debian & Ubuntu supported version have the 7 branch available:
https://apt.llvm.org/

We have now the package in the archive. Works fine on i386 and amd64.
Needs more time to investigate build results for the other archs:
We made some improvements to the llvm-toolchain packages in Debian & Ubuntu
blog post coming.

Besides that, for LLVM on Debian amd64:

  Expected Passes : 26034
  Expected Failures : 147
  Unsupported Tests : 810

For clang:

  Expected Passes : 12855
  Expected Failures : 19
  Unsupported Tests : 79
  Unexpected Failures: 21
(I think the failures are my fault)

And the rest seems fine.

Cheers,
Sylvestre