This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [v3] Use atomic builtins (powerpc, alpha)

>>>>> Paolo Carlini writes:

Paolo> Maybe better restating, more clearly, my point: in *my* understanding,
Paolo> maybe I'm wrong, the maintainers of the powerpc backend are happy with
Paolo> the current implementation of the atomic builtins. Therefore, we can use
Paolo> those builtins in the C++ runtime library as they are supposed to be
Paolo> used, to encapsulate completely those functionalities. Similarly to
Paolo> ia64: after all those builtins are modeled after the ia64 builtins.

Paolo> If you believe we are not really there, I'm ready to revert the powerpc
Paolo> bits. Just tell me, ok? And please accept my excuses if I'm mistaken in
Paolo> the general picture of the situation.

	The atomic builtins implemented by Geoff Keating are not a
complete replacement for the previous functionality.  Encapsulating
procesor-specific issues, like the PPC405 errata, in one location is
beneficial and all parts of the compiler should evolve to use the
builtins.  Very few places in the compiler have utilized atomic operations
until now: libstdc++ has been one such use.

	When one finds a special adaptation for a known problem, one
should not discard it without comment.  As Daniel mentioned, this should
be opened as a Critical Bugzilla target-specific bug that must be fixed
for GCC 4.1.0, at the very least.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]