This fix will be released in 1.10.1 in one week or two. 1.10.1 was supposed to happen one month ago but I am waiting for some other patches. If those are still delayed, I'll release 1.10.1 without them.
Can you tell me which test fails during make check on 32bit? My only failure here is in linux-libnuma but it's a i386/amd64 multiarch problem that's not worth fixing. Brice Le 21/11/2014 01:57, Thomas Van Doren a écrit : > Hi Brice > > Thank you for the quick response! That patch fixes the build issue and > hwloc works as expected (make check has 1 failure on 32bit, but that > also happens on master so I didn't worry about it). > > Will this fix be in the next hwloc release? If so, do you know > (approximately) when the next release will happen? > > Thank you! > > Thomas Van Doren > thomas.vando...@gmail.com <mailto:thomas.vando...@gmail.com> > > > On Wed, Nov 19, 2014 at 10:42 PM, Brice Goglin <brice.gog...@inria.fr > <mailto:brice.gog...@inria.fr>> wrote: > > Hello, > Thanks, I can reproduce the problem on Debian with -O3 -m32. > The issue is that -O3 makes gcc inline more. We have function A > call B multiple times, and B calls C which contains asm with a > label. So in the end A contains the asm label from C multiple times. > Google says we should use local labels to fix the confusion > between duplicates. > This patch should help. > Brice > > > --- a/include/private/cpuid-x86.h > +++ b/include/private/cpuid-x86.h > @@ -32,14 +32,14 @@ static __hwloc_inline int hwloc_have_x86_cpuid(void) > "pushfl \n\t" \ > "pop %1 \n\t" \ > "cmp %1,%2\n\t" /* Compare with expected value */ \ > - "jnz Lhwloc1\n\t" /* Unexpected, failure */ \ > + "jnz 0f\n\t" /* Unexpected, failure */ \ > > TRY_TOGGLE /* Try to set/clear */ > TRY_TOGGLE /* Try to clear/set */ > > "mov $1,%0\n\t" /* Passed the test! */ > > - "Lhwloc1: \n\t" > + "0: \n\t" > "popfl \n\t" /* Restore flags */ > > : "=r" (ret), "=&r" (tmp), "=&r" (tmp2)); > > > > > Le 20/11/2014 03:04, Thomas Van Doren a écrit : >> Hi >> >> I am getting the following error when building hwloc on 32bit >> ubuntu 14.10 (utopic) with optimizations (CFLAGS='-O3'): >> >> ... snip ... >> CC topology-x86.lo >> /home/vagrant/src/hwloc/include/private/cpuid-x86.h: Assembler >> messages: >> /home/vagrant/src/hwloc/include/private/cpuid-x86.h:40: Error: >> symbol `Lhwloc1' is already defined >> Makefile:878: recipe for target 'topology-x86.lo' failed >> make[1]: *** [topology-x86.lo] Error 1 >> make[1]: Leaving directory '/home/vagrant/src/hwloc/hwloc' >> Makefile:615: recipe for target 'all-recursive' failed >> make: *** [all-recursive] Error 1 >> >> I used these steps to verify that the build works without >> optimizations and reproduce the error above with optimizations: >> >> git clone https://github.com/open-mpi/hwloc >> cd hwloc/ >> ./autogen.sh >> ./configure >> make >> # This works, but no optimizations. >> >> # This fails with the error above... >> git clean -dxf >> ./autogen.sh >> ./configure CFLAGS='-O3' CXXFLAGS='-O3' >> make >> >> Is this expected to work? Do you have recommendations for fixing? >> I first saw this on hwloc 1.9.1 and verified it also happens on >> 1.10.0. >> >> I attached the configure and make V=1 output, as well as the >> config.log for the build against master (sha: 77fbe8f) that fails. >> >> I am using the utopic32 vagrant box from Ubuntu: >> http://cloud-images.ubuntu.com/vagrant/utopic/current/ >> >> Please let me know if you need any other information. >> >> Thanks! >> >> Thomas Van Doren >> thomas.vando...@gmail.com <mailto:thomas.vando...@gmail.com> >> >> >> _______________________________________________ >> hwloc-users mailing list >> hwloc-us...@open-mpi.org <mailto:hwloc-us...@open-mpi.org> >> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users >> Link to this post: >> http://www.open-mpi.org/community/lists/hwloc-users/2014/11/1117.php > > > _______________________________________________ > hwloc-users mailing list > hwloc-us...@open-mpi.org <mailto:hwloc-us...@open-mpi.org> > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users > Link to this post: > http://www.open-mpi.org/community/lists/hwloc-users/2014/11/1119.php > > > > > _______________________________________________ > hwloc-users mailing list > hwloc-us...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users > Link to this post: > http://www.open-mpi.org/community/lists/hwloc-users/2014/11/1119.php
