gmpbench: Ultra40 Solaris10 Studio11
tg-this-will-bounce-but-I-am-subscribed-to-the-list-honest at swox.com
Tue Apr 10 13:33:58 CEST 2007
Marc Glisse <marc.glisse+gmp at normalesup.org> writes:
> Why would one want to do that? It will not be used by longlong.h,
> since it is conditional on __GNUC__.
But then it is pretty trivial to add "|| defined __SUNPRO_C" to the lines
checking for __GNUC__ in longlong.h. What else would you advise for
someone who wants optimal performance using Sun's compiler?
We advice people that want optimal GMP performance to use GCC.
Trying to support inline asembly for non-free compilers, would mean
opening a can of worms. What subset of GCC's syntax is supported? It
is not really documented, so reverse engineering is the only option.
That requires access to many compiler version, since these things
change with version, of course.
Your suggestion to simply test __SUNPRO_C is certainly not right. It
would feed uncomprehensible code to older versons of the Sun compiler.
> Of course, Sun can do an Intel and have their compiler define
> __GNUC__, and cause complete confusion.
They don't intend to (they have said so pretty clearly).
More information about the gmp-devel