[libcxx-dev] [cfe-dev] [Release-testers] [8.0.0 Release] rc2 has been tagged

[Eli Friedman via libcxx-dev]

If each stage consistently produces the same object file, then treat it like any other miscompile: mix the object files created by a good/bad compiler to bisect which C++ file is getting miscompiled, then reduce the failure using opt-bisect-limit or bugpoint.  Or maybe you could try asan/ubsan first to rule out something silly.

If it’s non-determinism, you could try to use opt-bisect-limit to figure out which pass is non-deterministic.

-Eli

From: cfe-dev <cfe-dev-bounces at lists.llvm.org> On Behalf Of Brian Cain via cfe-dev
Sent: Tuesday, February 12, 2019 4:38 PM
To: Hans Wennborg <hans at chromium.org>
Cc: LLVM Development List <llvm-dev at lists.llvm.org>; Release-testers <release-testers at lists.llvm.org>; clangd-dev at lists.llvm.org; openmp-dev (openmp-dev at lists.llvm.org) <openmp-dev at lists.llvm.org>; LLDB Dev <lldb-dev at lists.llvm.org>; cfe-dev <cfe-dev at lists.llvm.org>; libcxx-dev at lists.llvm.org
Subject: [EXT] Re: [cfe-dev] [Release-testers] [8.0.0 Release] rc2 has been tagged

Any tips or suggestions on how to isolate or identify this defect?  It seems non-trivial to bisect this failure.

Hans: presumably this bug should block the 8.0 release?

On Mon, Feb 11, 2019 at 8:37 AM Brian Cain <brian.cain at gmail.com<mailto:brian.cain at gmail.com>> wrote:
rc1 did not exhibit this mismatch.  A repeat of the rc2 build repeated the mismatch.  I diff'd the disassembly between phase 2 and phase 3 and the difference is the same on both builds.  The difference follows:

# diff x86isel_p{2,3}.s
2c2
< Phase2/Release/llvmCore-8.0.0-rc2.obj/lib/Target/X86/CMakeFiles/LLVMX86CodeGen.dir/X86ISelLowering.cpp.o:     file format elf64-x86-64
---
> Phase3/Release/llvmCore-8.0.0-rc2.obj/lib/Target/X86/CMakeFiles/LLVMX86CodeGen.dir/X86ISelLowering.cpp.o:     file format elf64-x86-64
121757c121757
<      863:     48 8b 84 24 e0 05 00    mov    0x5e0(%rsp),%rax
---
>      863:     48 8b 84 24 b0 01 00    mov    0x1b0(%rsp),%rax
121759c121759
<      86b:     48 33 84 24 b0 01 00    xor    0x1b0(%rsp),%rax
---
>      86b:     48 33 84 24 e0 05 00    xor    0x5e0(%rsp),%rax
121761,121762c121761,121762
<      873:     8b 8c 24 e8 05 00 00    mov    0x5e8(%rsp),%ecx
<      87a:     33 8c 24 b8 01 00 00    xor    0x1b8(%rsp),%ecx
---
>      873:     8b 8c 24 b8 01 00 00    mov    0x1b8(%rsp),%ecx
>      87a:     33 8c 24 e8 05 00 00    xor    0x5e8(%rsp),%ecx


On Sat, Feb 9, 2019 at 9:54 AM Brian Cain <brian.cain at gmail.com<mailto:brian.cain at gmail.com>> wrote:
rc2 build on Ubuntu 16.04 saw a stage 2/3 mismatch on X86IselLowering.cpp.o -- I didn't try to build rc1 for 16.04 but I can try it now to see if it's there too.

I can pass along logs if anyone wants to investigate.

On Thu, Feb 7, 2019 at 9:41 AM Hans Wennborg via Release-testers <release-testers at lists.llvm.org<mailto:release-testers at lists.llvm.org>> wrote:
Dear testers,

8.0.0-rc2 has been tagged from the release_80 branch at r353413.

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

I'll get the source tarballs and docs published as soon as possible,
and binaries as they become available.

Thanks,
Hans
_______________________________________________
Release-testers mailing list
Release-testers at lists.llvm.org<mailto:Release-testers at lists.llvm.org>
https://lists.llvm.org/cgi-bin/mailman/listinfo/release-testers




--
-Brian


--
-Brian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/libcxx-dev/attachments/20190213/138c87e0/attachment-0001.html>