Re: bug in gmp on Opteron

[Greg Lindahl]

On second thought, it's not clear to me that this patch works. The gmp
source is somewhat confusingly arranged, and I think I ended up
testing the system library instead of the one I just made, and the one
I just made flunks the count_zeros testing.

> This patch works around it:
> 
> --- longlong.h~ 2002-12-17 13:57:20.000000000 -0800
> +++ longlong.h  2003-12-18 16:29:52.575408366 -0800
> @@ -715,8 +715,10 @@
>    } while (0)
>  #define count_trailing_zeros(count, x)                                 \
>    do {                                                                 \
> +    long __bug_temp;                                                    \
>      ASSERT ((x) != 0);                                                 \
> -    __asm__ ("bsfq %1,%0" : "=r" (count) : "rm" ((UDItype)(x)));       \
> +    __asm__ ("bsfq %1,%0" : "=r" (__bug_temp) : "rm" ((UDItype)(x)));  \
> +    count = __bug_temp;                                                 \
>    } while (0)
>  #endif /* x86_64 */