- ensure we don't build the 'context' and 'coroutine' libs, they are not
portable and the latter depends on the former
- don't use the 'gcc_sparc' backend in Boost.atomic, for it doesn't
build; use the generic gcc backend based on __sync_*
- stop tweaking cc -march / -mcpu values; this is not desirable on any
arch and it breaks on sparc64 due to a bogus -mcpu=c3 default
- don't attempt to force -m32 / -m64 flags, this breaks at least alpha
(64 bits but doesn't understand -m64)
alpha, hppa & sparc64 tests by landry@, amd64 bulk by ajacoutot@
ok jasper@
8495a25bd4