Compiler bugs for build-many-glibcs.py builds with GCC mainline

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Compiler bugs for build-many-glibcs.py builds with GCC mainline

Joseph Myers
FYI, the following failures building with build-many-glibcs.py with GCC
mainline (r242641, Sunday) appear to involve GCC bugs.  Hopefully some or
all of these will get fixed for GCC 7.

FAIL: compilers-m68k-linux-gnu-coldfire gcc-first build

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68467

FAIL: compilers-powerpc-linux-gnuspe gcc-first build

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78458

FAIL: compilers-tilegx-linux-gnu gcc-first build
FAIL: compilers-tilegxbe-linux-gnu gcc-first build

  Error building libgcc: "Assembler messages: Error: operand out of range
  (70 is not between 0 and 63)" (building _negvsi2.o).  Not yet filed in
  GCC Bugzilla.

FAIL: glibcs-microblaze-linux-gnu build
FAIL: glibcs-microblazeel-linux-gnu build

  Missing EH support in GCC, maintainers requested to get the required
  patch upstream.

FAIL: glibcs-sh3-linux-gnu check
FAIL: glibcs-sh3eb-linux-gnu check
FAIL: glibcs-sh4-linux-gnu-soft check
FAIL: glibcs-sh4-linux-gnu check
FAIL: glibcs-sh4eb-linux-gnu-soft check
FAIL: glibcs-sh4eb-linux-gnu check

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78459
  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78460

  (note the latter is cc1 eating all memory and swap until the OOM killer
  kills it, so watch out if using build-many-glibcs.py with GCC mainline)

--
Joseph S. Myers
[hidden email]
Reply | Threaded
Open this post in threaded view
|

Re: Compiler bugs for build-many-glibcs.py builds with GCC mainline

Jeff Law
On 11/21/2016 06:52 PM, Joseph Myers wrote:
> FYI, the following failures building with build-many-glibcs.py with GCC
> mainline (r242641, Sunday) appear to involve GCC bugs.  Hopefully some or
> all of these will get fixed for GCC 7.
>
> FAIL: compilers-m68k-linux-gnu-coldfire gcc-first build
>
>   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68467
If you could pass along a .i file for this it'd be appreciated.  I think
this is related to STACK_PREFERRED_BOUNDARY, but I'm not 100% sure and
the .i file would allow me to really dig into it.

jeff

Reply | Threaded
Open this post in threaded view
|

Re: Compiler bugs for build-many-glibcs.py builds with GCC mainline

Joseph Myers
On Tue, 22 Nov 2016, Jeff Law wrote:

> On 11/21/2016 06:52 PM, Joseph Myers wrote:
> > FYI, the following failures building with build-many-glibcs.py with GCC
> > mainline (r242641, Sunday) appear to involve GCC bugs.  Hopefully some or
> > all of these will get fixed for GCC 7.
> >
> > FAIL: compilers-m68k-linux-gnu-coldfire gcc-first build
> >
> >   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68467
> If you could pass along a .i file for this it'd be appreciated.  I think this
> is related to STACK_PREFERRED_BOUNDARY, but I'm not 100% sure and the .i file
> would allow me to really dig into it.

I've attached preprocessed source to that bug (along with giving compiler
options to reproduce the ICE and the configure options generated by
build-many-glibcs.py).  (Those are all generated with current GCC trunk,
but as noted the bug appears from 4.7 onwards.)

--
Joseph S. Myers
[hidden email]
Reply | Threaded
Open this post in threaded view
|

Re: Compiler bugs for build-many-glibcs.py builds with GCC mainline

Chris Metcalf-2
In reply to this post by Joseph Myers
On 11/21/2016 8:52 PM, Joseph Myers wrote:
> FAIL: compilers-tilegx-linux-gnu gcc-first build
> FAIL: compilers-tilegxbe-linux-gnu gcc-first build
>
>    Error building libgcc: "Assembler messages: Error: operand out of range
>    (70 is not between 0 and 63)" (building _negvsi2.o).  Not yet filed in
>    GCC Bugzilla.

This has been fixed in tilegx compiler backend and should be upstream
shortly.

--
Chris Metcalf, Mellanox Technologies
http://www.mellanox.com