[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#772436: marked as done (aarch64 optimization bug)

Your message dated Tue, 20 Sep 2016 22:15:11 +0000
with message-id <[🔎] E1bmTJn-0005QY-0D@franck.debian.org>
and subject line Bug#838316: Removed package(s) from unstable
has caused the Debian Bug report #772436,
regarding aarch64 optimization bug
to be marked as done.

This means that you claim that the problem has been dealt with.
If this is not the case it is now your responsibility to reopen the
Bug report if necessary, and/or fix the problem forthwith.

(NB: If you are a system administrator and have no idea what this
message is talking about, this may indicate a serious mail system
misconfiguration somewhere. Please contact owner@bugs.debian.org
immediately.)


-- 
772436: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=772436
Debian Bug Tracking System
Contact owner@bugs.debian.org with problems
--- Begin Message --- 
Package: gcc-4.9
Version: 4.9.1-19
Severity: normal

I have a minimal reproducable test case in a git repo here:

    https://github.com/erikd/gcc-aarch64-optimization-bug
    
which contails a test program which passes when run with
optimization flags "-O1" and fails with "-O1 -fschedule-insns".

Cloning that repo under aarch64 and doing:

    CC=gcc-4.9 make clean check
    
compiles and runs the program with both sets of optimization
flags and displays success or failure. It does not hwoever
fail on aarch64 with gcc-4.8 or with gcc-snapshot, nor does
it fail with any version of GCC on any other architecture.

There is a usefule comment from Måns Rullgård on G+:
    
    https://plus.google.com/u/0/+ErikdeCastroLopo/posts/F2BPxT7g1Gx
    
    OK, it's a gcc bug. For some unfathomable reason, it chooses to
    calculate "mask = ~0u >> (32 - numBits)" in a vector register by
    means of left-shifting by -(32 - numBits). It then subtracts 
    (bitPos & 7) from the negated value and uses this as the shift 
    amount of a regular shift instruction in "mask <<= shift" which
    is obviously nonsense.

    This behaviour (and the bug) was introduced in rev 202020 (git
    5674183), and reverting this commit makes things work again.

    I believe the -fschedule-insns flag is a red herring. It probably 
    merely stirs up the code in a way that makes the real bug trigger.

Older discussion about this problem here:

    https://plus.google.com/u/0/+ThomasRuecker/posts/e9uxoUyk3Bk

-- System Information:
Debian Release: 8.0
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: arm64 (aarch64)

Kernel: Linux 3.16.0-4-amd64 (SMP w/4 CPU cores)
Locale: LANG=en_AU.UTF-8, LC_CTYPE=en_AU.UTF-8 (charmap=UTF-8) (ignored: LC_ALL set to en_AU.UTF-8)
Shell: /bin/sh linked to /bin/dash
Init: unable to detect

Versions of packages gcc-4.9 depends on:
ii  binutils        2.24.90.20141201-1
ii  cpp-4.9         4.9.2-5
ii  gcc-4.9-base    4.9.2-5
ii  libc6           2.19-13
ii  libcloog-isl4   0.18.2-1+b2
ii  libgcc-4.9-dev  4.9.2-5
ii  libgmp10        2:6.0.0+dfsg-6
ii  libisl10        0.12.2-2+b2
ii  libmpc3         1.0.2-1+b2
ii  libmpfr4        3.1.2-1+b2
ii  zlib1g          1:1.2.8.dfsg-2+b1

Versions of packages gcc-4.9 recommends:
ii  libc6-dev  2.19-13

Versions of packages gcc-4.9 suggests:
pn  gcc-4.9-doc      <none>
pn  gcc-4.9-locales  <none>
pn  libasan1-dbg     <none>
pn  libatomic1-dbg   <none>
pn  libcilkrts5-dbg  <none>
pn  libgcc1-dbg      <none>
pn  libgomp1-dbg     <none>
pn  libitm1-dbg      <none>
pn  liblsan0-dbg     <none>
pn  libquadmath-dbg  <none>
pn  libtsan0-dbg     <none>
pn  libubsan0-dbg    <none>

-- no debconf information

--- End Message ---
--- Begin Message --- 
Version: 4.9.4-2+rm

Dear submitter,

as the package gcc-4.9 has just been removed from the Debian archive
unstable we hereby close the associated bug reports.  We are sorry
that we couldn't deal with your issue properly.

For details on the removal, please see https://bugs.debian.org/838316

The version of this package that was in Debian prior to this removal
can still be found using http://snapshot.debian.org/.

This message was generated automatically; if you believe that there is
a problem with it please contact the archive administrators by mailing
ftpmaster@ftp-master.debian.org.

Debian distribution maintenance software
pp.
Chris Lamb (the ftpmaster behind the curtain)

--- End Message ---
Reply to: