Re: [PATCH] Remove atomic-machine headers

[Florian Weimer <fweimer@redhat.com>]

* Wilco Dijkstra via Libc-alpha:

> diff --git a/configure b/configure
> index ff2c406b3b573484c843742d292fc2e0ee0e3008..3efdf45f99af71a482509f559d9de90aa294700c 100755
> --- a/configure
> +++ b/configure
> @@ -727,6 +727,7 @@ infodir
>  docdir
>  oldincludedir
>  includedir
> +runstatedir
>  localstatedir
>  sharedstatedir
>  sysconfdir

Spurious difference.

> +typedef struct { long long t; } X;
> +extern void has_64b_atomics(void);
> +void f(void)
> +{
> +  X x;
> +  /* Use address of structure with 64-bit type.  This avoids incorrect
> +     implementations which return true even if long long is not 64-bit aligned.
> +     This works on GCC and LLVM - other cases have bugs and they disagree.  */
> +  if (__atomic_always_lock_free (sizeof (x), &x))
> +    has_64b_atomics();

Can you use _static_assert, given that it's supposed to yield a
compile-time constant?

I'd be suprised if __atomic_always_lock_free is correct on all
glibc-supported architectures, though.

> diff --git a/sysdeps/x86/atomic-machine.h b/sysdeps/x86/atomic-machine.h
> index 98541a2d06ff5e4aa8c789ab7405215097471971..303458d1296e4b1cd7cd654e0904ace0ffc52fae 100644
> --- a/sysdeps/x86/atomic-machine.h
> +++ b/sysdeps/x86/atomic-machine.h
> @@ -21,17 +21,6 @@
>  
>  #include <tls.h>			/* For mach.  */
>  
> -#ifdef __x86_64__
> -# define __HAVE_64B_ATOMICS		1
> -#else
> -/* Since the Pentium, i386 CPUs have supported 64-bit atomics, but the
> -   i386 psABI supplement provides only 4-byte alignment for uint64_t
> -   inside structs, so it is currently not possible to use 64-bit
> -   atomics on this platform.  */
> -# define __HAVE_64B_ATOMICS		0
> -#endif
> -#define ATOMIC_EXCHANGE_USES_CAS	0
> -
>  #define atomic_spin_nop() __asm ("pause")
>  
>  #endif /* atomic-machine.h */

Please preserve this comment somewhere.  Maybe folod it into the
configure test example.

Thanks,
Florian