Re: [PATCH v2] For RTEMS-LIBBSD support, add bitcount routines

[Sebastian Huber <sebastian.huber@embedded-brains.de>]

On 17/05/17 15:06, Kevin Kirspel wrote:
> ---
>   newlib/libc/sys/rtems/include/machine/types.h | 58 +++++++++++++++++++++++++++
>   1 file changed, 58 insertions(+)
>
> diff --git a/newlib/libc/sys/rtems/include/machine/types.h b/newlib/libc/sys/rtems/include/machine/types.h
> index b28f923..0b0a100 100644
> --- a/newlib/libc/sys/rtems/include/machine/types.h
> +++ b/newlib/libc/sys/rtems/include/machine/types.h
> @@ -83,3 +83,61 @@ typedef	__uintmax_t	rman_res_t;
>   #endif
>   
>   #endif /* __BSD_VISIBLE */
> +
> +#ifdef __POPCNT__
> +#define	__bitcount64(x)	__builtin_popcountll((__uint64_t)(x))
> +#define	__bitcount32(x)	__builtin_popcount((__uint32_t)(x))
> +#define	__bitcount16(x)	__builtin_popcount((__uint16_t)(x))
> +#define	__bitcountl(x)	__builtin_popcountl((unsigned long)(x))
> +#define	__bitcount(x)	__builtin_popcount((unsigned int)(x))
> +#else /* __POPCNT__ */
> +/*
> + * Population count algorithm using SWAR approach
> + * - "SIMD Within A Register".
> + */
> +static __inline __uint16_t
> +__bitcount16(__uint16_t _x)
> +{
> +	_x = (_x & 0x5555) + ((_x & 0xaaaa) >> 1);
> +	_x = (_x & 0x3333) + ((_x & 0xcccc) >> 2);
> +	_x = (_x + (_x >> 4)) & 0x0f0f;
> +	_x = (_x + (_x >> 8)) & 0x00ff;
> +	return (_x);
> +}
> +
> +static __inline __uint32_t
> +__bitcount32(__uint32_t _x)
> +{
> +	_x = (_x & 0x55555555) + ((_x & 0xaaaaaaaa) >> 1);
> +	_x = (_x & 0x33333333) + ((_x & 0xcccccccc) >> 2);
> +	_x = (_x + (_x >> 4)) & 0x0f0f0f0f;
> +	_x = (_x + (_x >> 8));
> +	_x = (_x + (_x >> 16)) & 0x000000ff;
> +	return (_x);
> +}
> +
> +#if __SIZEOF_INT__ == 8
> +static __inline __uint64_t
> +__bitcount64(__uint64_t _x)
> +{
> +	_x = (_x & 0x5555555555555555) + ((_x & 0xaaaaaaaaaaaaaaaa) >> 1);
> +	_x = (_x & 0x3333333333333333) + ((_x & 0xcccccccccccccccc) >> 2);
> +	_x = (_x + (_x >> 4)) & 0x0f0f0f0f0f0f0f0f;
> +	_x = (_x + (_x >> 8));
> +	_x = (_x + (_x >> 16));
> +	_x = (_x + (_x >> 32)) & 0x000000ff;
> +	return (_x);
> +}
> +
> +#define	__bitcountl(x)	__bitcount64((unsigned long)(x))
> +#else /* __SIZEOF_INT__ == 8 */
> +static __inline __uint64_t
> +__bitcount64(__uint64_t _x)
> +{
> +	return (__bitcount32(_x >> 32) + __bitcount32(_x));
> +}

The __bitcount64() implementation should be based on the word size of 
the architecture. Usually this is indicated by sizeof long.

> +
> +#define	__bitcountl(x)	__bitcount32((unsigned long)(x))
> +#endif /* __SIZEOF_INT__ == 8 */
> +#define	__bitcount(x)	__bitcount32((unsigned int)(x))
> +#endif /* __POPCNT__ */

Selection of __bitcountl() should be based on sizeof long. Selection of 
__bitcount() should be based on sizeof int. Maybe it makes sense to add 
a test for this to the RTEMS testsuite and do a check of the 
pre-processor output on the arm-rtems4.12, sparc64-rtems4.12 and 
m32c-rtems4.12 targets.

-- 
Sebastian Huber, embedded brains GmbH

Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone   : +49 89 189 47 41-16
Fax     : +49 89 189 47 41-09
E-Mail  : sebastian.huber@embedded-brains.de
PGP     : Public key available on request.

Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.