On Dec 1, 2:10 pm, François Bissey <f.r.bis...@massey.ac.nz> wrote: > On Wed, 02 Dec 2009 04:17:50 strogdon wrote: > > > Hey Francois. I too thought that perhaps the spkg supplied flags would > > override any changes with custom flags. However the custom flags seem > > to be appended to the spkg provided ones. For example > > > with custom CFLAGS > > building 'sage.algebras.quatalg.quaternion_algebra_element' extension > > gcc -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -march=opteron - > > O2 -pipe -fPIC > > > with CFLAGS unset > > building 'sage.algebras.quatalg.quaternion_algebra_element' extension > > gcc -fno-strict-aliasing -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict- > > prototypes -fPIC > > > The no-strict-aliasing get's lost! I'll give your suggestion a whirl, > > but it will take some time to rebuild things. > > I think that may be a scons behaviour. Yes I think I have seen some > stuff they do in polybori. That may explain the duplication of a number > of flags there. > -fno-strict-aliasing disappearance and appearance are strange. > I am not a scons specialist but I am guessing it is not set as intended. > > Please keep us up to date with the results. > Francois
Here are the results of the tests. When CFLAGS were passed to the build of sage they were passed as CFLAGS="..." CXXFLAGS="..." make ALL tests failed to build sage correctly except in the situations I listed previously when the CFLAGS were unset in spkg-install and when no CFLAGS were passed. In this latter case a simple 'make' was issued. All failures were identical to the failure listed previously. Below I list the flags that are passed to gcc in a typical build of the sage spkg components for each test. no CFLAGS a simple 'make' issued building 'sage.algebras.quatalg.quaternion_algebra_element' extension gcc -fno-strict-aliasing -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict- prototypes -fPIC CFLAGS="-O2 -pipe" building 'sage.algebras.quatalg.quaternion_algebra_element' extension gcc -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -O2 -pipe -fPIC CFLAGS="-pipe" building 'sage.algebras.quatalg.quaternion_algebra_element' extension gcc -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -pipe -fPIC CFLAGS="" and CFLAGS= (I find this one curious!) building 'sage.algebras.quatalg.quaternion_algebra_element' extension gcc -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC CFLAGS="-march=native" building 'sage.algebras.quatalg.quaternion_algebra_element' extension gcc -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -march=native - fPIC So, it would appear that it is not the CFLAGS but whether CFLAGS has been set to something. The no-strict-aliasing is suspiciously missing. The documentation I have on gcc 4.3.4 indicates that optimization levels O2 and O3 turn on strict-aliasing. Therefore if no-strict- aliasing is needed to built sage spkg then this may be the problem. But why for amd64 only? Steve -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
