Re: clang build buggy code with certain CPUTYPE setting

On Sat, Sep 26, 2020 at 05:48:33PM +0200, Dimitry Andric wrote:
> On 26 Sep 2020, at 13:40, Marek Zarychta <zarychtam_at_plan-b.pwste.edu.pl> wrote:
> > 
> > I have done a few builds of CURRENT in a row one or two weeks apart. The
> > builds with CPUTYPE?=amdfam10 set produce buggy code, for example while
> > running mergemaster I get this error:
> > 
> > PLEASE submit a bug report to https://bugs.freebsd.org/submit/ and
> > include the crash backtrace, preprocessed source, and associated run
> > script.
> > Stack dump:
> > 0.      Program arguments: cc --version
> > #0 0x00000000040ede6e (/usr/bin/cc+0x40ede6e)
> > #1 0x00000000040ec0e5 (/usr/bin/cc+0x40ec0e5)
> > #2 0x00000000040ee550 (/usr/bin/cc+0x40ee550)
> > #3 0x000000080553babe (/lib/libthr.so.3+0x19abe)
> > Illegal instruction
> > make: "/usr/src/share/mk/bsd.compiler.mk" line 181: Unable to determine
> > compiler type for CC=cc.  Consider setting COMPILER_TYPE.
> > 
> > The 13-CURRENT world built without CPUTYPE runs fine, the same for
> > recent 12.2-STABLE world build with CPUTYPE?=amdfam10 on the same
> > machine.
> 
> Hi Marek,
> 
> In r365507 (on 2020-09-09) I committed a fix for amdfam10:
> 
> ------------------------------------------------------------------------
> r365507 | dim | 2020-09-09 20:11:04 +0200 (Wed, 09 Sep 2020) | 17 lines
> 
> Merge commit e6bb4c8e7 from llvm git (by Craig Topper):
> 
>   [X86] SSE4_A should only imply SSE3 not SSSE3 in the frontend.
> 
>   SSE4_1 and SSE4_2 due imply SSSE3. So I guess I got confused when
>   switching the code to being table based in D83273.
> 
>   Fixes PR47464
> 
> This should fix builds with -march=amdfam10 emitting SSSE3 instructions
> such as pshufb, which lead to programs crashing with SIGILL on such
> processors.
> 
> Reported by:    avg
> MFC after:      6 weeks
> X-MFC-With:     r364284
> 
> So I expect that the "Illegal instruction" you are seeing is an SSSE3
> instruction. If this happens with your base compiler, please get a
> known-good copy from one of the snapshot images. Ensure your /usr/src is
> r365507 or later, then do a full buildworld and reinstall.
> 
> -Dimitry
> 
Dear Dimitry,

Thank you for the information and for the fix. Sadly I must admit it
doesn't work for me. I have tried two builds with fresh sources today to
be certain and it looks like the bug is still present on FreeBSD
13-CURRENT r366186. Either the upstream fixed it only partially or it is
another bug. As a workaround, I will build worlds without
CPUTYPE?=amdfam10 for a while. I hope the problem will be resolved
before clang 11 is MFCed to 12-STABLE.

-- 
Marek Zarychta