Re: [PATCH] arm: provide a_ctz_l and a_ctz_64 helper functions

[Rich Felker <dalias@libc.org>]

On Thu, Nov 30, 2017 at 04:00:12PM -0800, Andre McCurdy wrote:
> Provide an ARM specific a_ctz_l helper function for architecture
> versions for which it can be implemented efficiently via the "rbit"
> instruction (ie all Thumb-2 capable versions of ARM v6 and above).
> 
> Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
> ---
>  arch/arm/atomic_arch.h | 22 ++++++++++++++++++++++
>  1 file changed, 22 insertions(+)
> 
> diff --git a/arch/arm/atomic_arch.h b/arch/arm/atomic_arch.h
> index 6e2e3b4..9242df8 100644
> --- a/arch/arm/atomic_arch.h
> +++ b/arch/arm/atomic_arch.h
> @@ -91,4 +91,26 @@ static inline int a_clz_32(uint32_t x)
>  	return x;
>  }
>  
> +#if __ARM_ARCH_6T2__ || __ARM_ARCH_7A__ || __ARM_ARCH_7R__ || __ARM_ARCH >= 7
> +
> +#define a_ctz_l a_ctz_l
> +static inline int a_ctz_l(unsigned long x)
> +{
> +	uint32_t xr;
> +	__asm__ ("rbit %0, %1" : "=r"(xr) : "r"(x));
> +	return a_clz_32(xr);
> +}
> +
> +#define a_ctz_64 a_ctz_64
> +static inline int a_ctz_64(uint64_t x)
> +{
> +	uint32_t y = x;
> +	if (!y) {
> +		y = x>>32;
> +		return 32 + a_ctz_l(y);
> +	}
> +	return a_ctz_l(y);
> +}
> +
> +#endif
>  #endif
> -- 
> 1.9.1

Looks mostly good.

I realize it's already done this way some other places, but I think it
would make more sense to adjust src/internal/atomic.h to use a_ctz_l
to define a_ctz_64, using the above, in cases where the arch defines
the former but not the latter. This would eliminate the dupliation
across atomic_arch.h files.

Note that we can also provide ctz without rbit on older arm, via
31-a_clz_32(x&-x). This could probably also be done in
src/internal/atomic.h, but I'm not sure it helps since ARM is probably
the only arch with a clz but not ctz.

Rich