[Bug middle-end/50770] New: limits-fndefn.c times out on ARM; runs 22 x faster with -fno-var-tracking

[Bug middle-end/50770] New: limits-fndefn.c times out on ARM; runs 22 x faster with -fno-var-tracking

[michael.hope at linaro dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50770

             Bug #: 50770
           Summary: limits-fndefn.c times out on ARM; runs 22 x faster
                    with -fno-var-tracking
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: michael.hope@linaro.org


On a Cortex-A9 with recent trunk:

cbuild@tcpanda01:~$ time
./gcc-4.7~svn179959-armv7l-natty-cbuild199-tcpanda06-cortexa9r1/bin/gcc -O3 -g 
-c limits-fndefn.c 

real    17m44.543s
user    7m7.477s
sys    10m33.117s

Adding -fno-var-tracking gives:

cbuild@tcpanda01:~$ time
./gcc-4.7~svn179959-armv7l-natty-cbuild199-tcpanda06-cortexa9r1/bin/gcc -O3 -g
-fno-var-tracking -c limits-fndefn.c 

real    0m31.788s
user    0m30.398s
sys    0m0.937s

The times are similar on the 4.6 branch.

This isn't causing problems, but suggests that the var tracking code could be
improved.

[Bug middle-end/50770] limits-fndefn.c times out on ARM; runs 22 x faster with -fno-var-tracking

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50770

Steven Bosscher <steven at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-10-18
     Ever Confirmed|0                           |1

--- Comment #1 from Steven Bosscher <steven at gcc dot gnu.org> 2011-10-18 22:06:00 UTC ---
Known issue. Push Alex to work on it if his patch doesn't help enough:
http://gcc.gnu.org/ml/gcc-patches/2011-10/msg01641.html

[Bug middle-end/50770] limits-fndefn.c times out on ARM; runs 22 x faster with -fno-var-tracking

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50770

--- Comment #2 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2011-10-25 04:38:34 UTC ---
Michael, please fill in the target and any other configure options you used,
and attach a preprocessed version of the source that exposes the problem, if
the patch in r180194 didn't take care of it.  Thanks in advance,

[Bug middle-end/50770] limits-fndefn.c times out on ARM; runs 22 x faster with -fno-var-tracking

[michael.hope at linaro dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50770

Michael Hope <michael.hope at linaro dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |arm

--- Comment #3 from Michael Hope <michael.hope at linaro dot org> 2011-10-25 23:05:46 UTC ---
Added target.

The logs are available at:

http://builds.linaro.org/toolchain/gcc-4.7~svn179959/logs/armv7l-natty-cbuild199-tcpanda06-cortexa9r1/

GCC was configured with:

--prefix=/scratch/cbuild/slave/slaves/tcpanda06/gcc-4.7~svn179959/gcc/default/install
--enable-languages=c,c++,objc,obj-c++,fortran,lto --with-mode=thumb
--with-arch=armv7-a --with-tune=cortex-a9 --with-float=softfp --with-fpu=neon

More information on the host is at:

http://builds.linaro.org/toolchain/gcc-4.7~svn179959/logs/armv7l-natty-cbuild199-tcpanda06-cortexa9r1/host.txt

The test case is gcc/testsuite/gcc.c-torture/compile/limits-fndefn.c.

I'm building r180460 at the moment and will see how that goes.

[Bug middle-end/50770] limits-fndefn.c times out on ARM; runs 22 x faster with -fno-var-tracking

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=50770

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |danglin at gcc dot gnu.org

--- Comment #4 from John David Anglin <danglin at gcc dot gnu.org> ---
With 5.0, there is a segmentation fault on hppa64-hp-hpux11.11:

spawn /test/gnu/gcc/objdir/gcc/xgcc -B/test/gnu/gcc/objdir/gcc/
-fno-diagnostics
-show-caret -fdiagnostics-color=never -O2 -flto -flto-partition=none -w -c -o
li
mits-fndefn.o
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/compile/limits-fndef
n.c
^M
Pid 16274 received a SIGSEGV for stack rowth failure.^M
Possible causes: insufficient memory or swap space,^M
or stack size exceeded maxssiz. ^M
xgcc: internal compiler error: Segmentation fault (program cc1)

Stack size is 16384 kbytes.  This doesn't occur with 4.9.