Re: [PATCH] page: get_order() optimization

[H. Peter Anvin]

On 04/01/2011 12:18 PM, Maksym Planeta wrote:
> Loop was repalaced with __builtin_clz(). This still allows to precompute
> constants, but on some architectures it uses special instruction to
> calculate order.
> 
> Signed-off-by: Maksym Planeta <mcsim.planeta@xxxxxxxxx>
> ---
>  include/asm-generic/getorder.h |    8 +++-----
>  1 files changed, 3 insertions(+), 5 deletions(-)
> 
> diff --git a/include/asm-generic/getorder.h b/include/asm-generic/getorder.h
> index 67e7245..fe8020c 100644
> --- a/include/asm-generic/getorder.h
> +++ b/include/asm-generic/getorder.h
> @@ -11,11 +11,9 @@ static inline __attribute_const__ int get_order(unsigned long size)
>  	int order;
>  
>  	size = (size - 1) >> (PAGE_SHIFT - 1);
> -	order = -1;
> -	do {
> -		size >>= 1;
> -		order++;
> -	} while (size);
> +	order = (__builtin_clzl(size) ^ (BITS_PER_LONG - 1));
> +	if (size == 0)
> +		order = 0;
>  	return order;
>  }
>  

You need to guard this with __GNUC__ >= 4; there are still laggards
using gcc 3.  Furthermore, on some platforms __builtin_clz*() does a
libgcc call which may be undesirable.

For the generic case, one can do something like this instead of a loop:

static inline unsigned int __clzl(unsigned long v)
{
        unsigned int p;

#if BITS_PER_LONG == 64
	p = 63;

	if (v & 0xffffffff00000000UL) {
		p -= 32;
		v >>= 32;
	}
#else
	p = 31;
#endif

        if (v & 0xffff0000) {
                p -= 16;
                v >>= 16;
        }
        if (v & 0xff00) {
                p -= 8;
                v >>= 8;
        }
        if (v & 0xf0) {
                p -= 4;
                v >>= 4;
        }
        if (v & 0xc) {
                p -= 2;
                v >>= 2;
        }
        if (v & 0x2) {
                p -= 1;
                v >>= 1;
        }

        return p;
}
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/