Stewart Smith wrote:
> On Thu, Apr 09, 2009 at 12:50:48AM +0200, Paul McCullagh wrote:
>> bzr branch lp:~drizzle-pbxt/drizzle/drizzle-pbxt
>
> Getting link failures on linux/sparc (you have ssh to the box, so feel
> free to try there if you like):
>
> ../storage/pbxt/src/libpbxt.a(libpbxt_a-lock_xt.o): In function
> `xt_atomic_dec2(unsigned short volatile*)':
> /home/stewart/drizzle/drizzle-pbxt/storage/pbxt/src/lock_xt.h:209:
> undefined reference to `__sync_fetch_and_sub_2'
> ../storage/pbxt/src/libpbxt.a(libpbxt_a-lock_xt.o): In function
> `xt_atomic_inc2(unsigned short volatile*)':
> /home/stewart/drizzle/drizzle-pbxt/storage/pbxt/src/lock_xt.h:194:
> undefined reference to `__sync_fetch_and_add_2'
> collect2: ld returned 1 exit status
>
> IIRC mtaylor had similar he had to sort out with the GCC atomics work.
Yes... there are some atomic issues. This:
#elif defined(__GNUC__)
__sync_fetch_and_add(mptr, 1);
is not valid, as that is not in existence everywhere (like sparc)
> You can probably find the right SPARC assembler for it pretty easily
> though if that'd help.
>
> on a slightly different note, for the default #else there, The ++ and --
> operators I don't think are excellent choices... going instead for
> #error or a lock would be better (and produce correct results).
In drizzled/atomic.h we fall back to pthread locks on platforms that
don't have TBB or gcc atomics or Sun atomics. You're welcome to use
drizzled/atomic.h in your code, of course... :)
Monty
_______________________________________________
Mailing list: https://launchpad.net/~drizzle-discuss
Post to : [email protected]
Unsubscribe : https://launchpad.net/~drizzle-discuss
More help : https://help.launchpad.net/ListHelp
