I've tagged LLVM 11.0.1-rc1. Testers may begin testing and uploading binaries. If you still have bugs you want fixed in LLVM 11.0.1, you have until Dec. 8 to request backports. You can make these requests by filing a bug at bugs.llvm.org and putting release-11.0.1 in the 'blocks' field.
Windows is ready:
$ sha256sum LLVM-11.0.1-rc1-win*.exe
They were built using the attached script.
build_llvm_1101-rc1.bat|attachment (4.95 KB)
Uploaded Ubuntu 20.10.
FAIL: SanitizerCommon-Unit :: ./Sanitizer-x86_64-Test/SanitizerLinux.ThreadDescriptorSize (66416 of 71964)
******************** TEST 'SanitizerCommon-Unit :: ./Sanitizer-x86_64-Test/SanitizerLinux.ThreadDescriptorSize' FAILED ********************
Note: Google Test filter = SanitizerLinux.ThreadDescriptorSize
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from SanitizerLinux
[ RUN ] SanitizerLinux.ThreadDescriptorSize
Which is: 2496
To be equal to: ThreadDescriptorSize()
Which is: 2304
[ FAILED ] SanitizerLinux.ThreadDescriptorSize (10 ms)
[----------] 1 test from SanitizerLinux (10 ms total)
[----------] Global test environment tear-down
[==========] 1 test from 1 test case ran. (10 ms total)
[ PASSED ] 0 tests.
[ FAILED ] 1 test, listed below:
[ FAILED ] SanitizerLinux.ThreadDescriptorSize
1 FAILED TEST
Testing: 0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90..
Failed Tests (1):
SanitizerCommon-Unit :: ./Sanitizer-x86_64-Test/SanitizerLinux.ThreadDescriptorSize
Testing Time: 338.33s
Unsupported : 1766
Passed : 69949
Expectedly Failed: 248
Failed : 1
/home/nnelson/Documents/llvm-project/llvm/utils/release/rc1/llvm-test-suite/MultiSource/Benchmarks/DOE-ProxyApps-C++/CLAMR/memstats.c:60:10: fatal error: 'sys/sysctl.h' file not found
When running ``llvm-test-suite
/home/nnelson/Documents/llvm-project/llvm/utils/release/rc1/llvm-test-suite/MultiSource/Benchmarks/DOE-ProxyApps-C++/CLAMR/memstats.c:60:10: fatal error: ‘sys/sysctl.h’ file not found
[ 30%] Linking C executable lpbench
1 error generated.
make: *** [MultiSource/Benchmarks/DOE-ProxyApps-C++/CLAMR/CMakeFiles/CLAMR.dir/build.make:362: MultiSource/Benchmarks/DOE-ProxyApps-C++/CLAMR/CMakeFiles/CLAMR.dir/memstats.c.o] Error 1
For 11.0.1 rc1, I have built and tested on both FreeBSD 11 and 12 this time. No additional patches were needed.
Main results on amd64-freebsd11:
Unsupported : 5122 (11.0.0 final: 5122)
Passed : 69779 (11.0.0 final: 69765)
Expectedly Failed : 245 (11.0.0 final: 245)
Timed Out : 16 (11.0.0 final: 16)
Failed : 478 (11.0.0 final: 479)
Unexpectedly Passed: 2 (11.0.0 final: 2)
Test suite results on amd64-freebsd11:
Passed: 2400 (11.0.0 final: 2400)
Failed: 2 (11.0.0 final: 2)
Main results on amd64-freebsd12:
Unsupported : 5122
Passed : 69776
Expectedly Failed : 245
Timed Out : 16
Failed : 481
Unexpectedly Passed: 2
Test suite results on amd64-freebsd12:
Main results on i386-freebsd11:
Unsupported : 3512
Passed : 66597
Expectedly Failed : 230
Timed Out : 7
Failed : 322
Unexpectedly Passed: 1
SHA256 (clang+llvm-11.0.1-rc1-amd64-unknown-freebsd11.tar.xz) = 92d956dbe14176c8defeb32c03449ed61a42eb0ba0b500784fff6f8a62b00f01
SHA256 (clang+llvm-11.0.1-rc1-amd64-unknown-freebsd12.tar.xz) = 7d38dcbe47d62da3dd555516ed31e5749edb952163e9c66cc3b12926a78e8f88
SHA256 (clang+llvm-11.0.1-rc1-i386-unknown-freebsd11.tar.xz) = 2e80d8d6c7178374696c67e2fea6a7d9cadf68bca08e07238b814d46cea09fa3
SHA256 (clang+llvm-11.0.1-rc1-i386-unknown-freebsd12.tar.xz) = b3d4ca7539f3e29a47b8f724503223051e6a62157b966afa6a4aba9a89400ef7
is there a place to download a LLVM clang which target IBM AIX PowerPC?
MacOS build: clang+llvm-11.0.1-rc1-x86_64-apple-darwin.tar.xz
with SHA256: c9ee87d7e42df8494a9f42993ed499479b3ce118c940a6e8907d075ceb913223
The same tests as before failed:
FAIL: libunwind :: libunwind_01.pass.cpp (69255 of 69302)
FAIL: libunwind :: signal_frame.pass.cpp (69258 of 69302)
I had to use the following patch to use Python 3:
diff --git a/openmp/runtime/test/lit.cfg b/openmp/runtime/test/lit.cfg
index 357b18a205d..96c0c3a1da7 100644
@@ -76,7 +76,7 @@ if config.operating_system == 'Darwin':
cmd = subprocess.Popen(['xcrun', '--show-sdk-path'],
out, err = cmd.communicate()
- out = out.strip()
+ out = out.strip().decode()
res = cmd.wait()
if res == 0 and out:
config.test_flags += " -isysroot " + out
otherwise tests failed to run.
I just realized that we should probably make an arm64 build as well
for those new fancy mac's.
What do people think is the best solution here, a fat universal build
that will be double the size (currently llvm+clang is already ~400MB
packed so it will be very big), or two separate builds?
The pro of having two separate builds is that we can set the correct
default triple instead of a single one, the downside is that I need to
build it twice for every version.
On Windows we're building two different packages (32-bit and 64-bit) in the same script , perhaps the same should be done here? Smaller packages is better?
I think the separate packages make the most sense.
Also, how would one practically go about doing the fat binary
approach? Is there some cmake magic that would make it
Yeah you can pass -DCMAKE_OSX_ARCHITECTURES=arm64;x86_64 and it will
make fat binaries.
But it seems like we should probably do two packages. That probably
needs to be implemented in the test-release in the following way:
Build Phase1 for the host currently running on. Then build Phase2 and
3 for the target (arm64) and compare those.
Anything I am not thinking about here or missing?
Hmm, but unless you're doing this on an arm64 machine, you won't be
able to run the tests in Phase2 and 3?
Thats a very good point.
Maybe someone with an M1 machine can take on the task of testing releases on those? I would happily do that but I don’t have such a machine and can’t afford one right now unfortunately.
Nico has an M1 and I think he looked at building LLVM on it too.
Perhaps he can help out.
Sure, can do. I know that tests don’t currently pass on M1s on trunk:
check-lld is fine, haven’t tried anything else yet. What would the goal be? Make sure the 11.0 branch isn’t worse than trunk?
Yeah I think run the test-release script and if it's not way worse
than on master we can upload it as part of the release. If it's
unusable in the current state we should probably skip 11.0.1 and wait
until we have landed some fixes in 12.
I finished testing llvm-11.0.1-rc1 on Power PC 64bit Little Endian for
- Ubuntu 18.04
- Red Hat 7.4
I have uploaded the binaries from IBM. The sha1 files are attached.
clang+llvm-11.0.1-rc1-powerpc64le-linux-rhel-7.4.sha1 (98 Bytes)
clang+llvm-11.0.1-rc1-powerpc64le-linux-ubuntu-18.04.sha1 (102 Bytes)
I've updated OpenMandriva and rebuilt all packages, no regressions from 11.0.0 detected on all arches.
I’ve uploaded binaries for armv7 & aarch64. No regressions compared to 11.0.0.
Uploaded Ubuntu 16 tarball:
BTW - Neil – if you decide to post binaries for Ubuntu 20, you should know that there was a request in another thread to include LLDB in the ubuntu packages. The ubuntu binaries I’ve made do include it (test-release takes an “-lldb” arg).
IBM has been working towards enabling LLVM development for AIX with the LLVM community. Currently, there are several OS header updates that IBM needs to deploy with the OS in order to work correctly with clang. I hope to publish a package, which includes clang binaries for AIX so that everyone has access to a build compiler for LLVM on AIX, though it’s not available yet. If an early release of such a package would be useful to you, please let me know and I’ll see if I can cut some red tape and get you something. There are some significant limitations to these compilers. These include challenges with functional exception handling, altivec, debug support, and TLS, to name a few, nonetheless, we have been able to bootstrap LLVM on AIX with some temporary downstream work.