Re: __sync_lock_test_and_set on ARM

[David Daney <ddaney@xxxxxxxxxx>]

Phil Endecott wrote:
> David Daney wrote:
> > > > > > The proper symbol to test would be something like: 
> > > > > > __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 or 
> > > > > > __GCC_HAVE_SYNC_LOCK_TEST_AND_SET_4  they are automatically set by 
> > > > > > GCC when the builtin functions are available.
>
> > > > $ /home/daney/gccsvn/mips64-trunk/gcc/xgcc 
> > > > -B/home/daney/gccsvn/mips64-trunk/gcc/ -E -dM st.c | grep SYNC
> > > > #define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 1
> > > > #define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8 1
>
> Looking at 
> http://gcc.gnu.org/viewcvs/trunk/gcc/c-cppbuiltin.c?revision=128218&view=markup 
>
> it seems that the __GCC_HAVE_SYNC_COMPARE_AND_SWAP_n macros are 
> explicitly defined if the machine supports them, but not any of the 
> other __SYNC_something builtins.  Unless they come from somewhere else.

They don't exist.  I think the reasoning is as follows:  If the 
compare_and_swap builtin exists, then all others can (and will) be 
synthesized from it if they do not have their own implementations.  So 
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_n indicates that  *all* of the 
primitives exist.

David Daney