Re: [PATCH v2 08/30] Add sparc-specific parity functions

[Sam Ravnborg <sam@ravnborg.org>]

Hi Zeng.

> 
> Use runtime patching for sparc64, lifted from hweight
No errors found in patch - but a few comments.
In general patch looks good.

> +++ b/arch/sparc/include/asm/bitops_64.h
> @@ -47,6 +47,24 @@ unsigned int __arch_hweight16(unsigned int w);
>  unsigned int __arch_hweight8(unsigned int w);
>  
>  #include <asm-generic/bitops/const_hweight.h>
> +
> +/*
> + * parityN: returns the parity of a N-bit word,
> + * i.e. the number of 1-bits in w modulo 2.
> + */
> +
> +static inline unsigned int __arch_parity4(unsigned int w)
> +{
> +	w &= 0xf;
> +	return (0x6996 >> w) & 1;
> +}
As Josef already said - this constant should have a name.
PARITY_BIT
?

> +++ b/arch/sparc/kernel/sparc_ksyms_64.c
> @@ -45,6 +45,12 @@ EXPORT_SYMBOL(__arch_hweight16);
>  EXPORT_SYMBOL(__arch_hweight32);
>  EXPORT_SYMBOL(__arch_hweight64);
>  
> +/* from parity.S */
> +EXPORT_SYMBOL(__arch_parity8);
> +EXPORT_SYMBOL(__arch_parity16);
> +EXPORT_SYMBOL(__arch_parity32);
> +EXPORT_SYMBOL(__arch_parity64);

Did you compile this?
I wonder if bitops_64.h is indirectly included.

> index 0000000..b1945e3
> --- /dev/null
> +++ b/arch/sparc/lib/parity.S
> @@ -0,0 +1,93 @@
> +#include <linux/linkage.h>
> +
> +	.text
> +	.align	32
> +
> +ENTRY(__arch_parity8)
> +	srl		%o0, 4, %g1
> +	xor		%o0, %g1, %o0
> +	and		%o0, 0xf, %o0
> +	sethi		%hi(0x6996), %g1
> +	or		%g1, %lo(0x6996), %g1
> +	srl		%g1, %o0, %o0
> +	retl
> +	 and		%o0, 1, %o0
> +ENDPROC(__arch_parity8)

I know the level of comments in hweight is equal to none.
But please do not follow this bad example.
At least for each function a one-liner of the C code.
And in the top of the file maybe two lines that the functions
are patched at run-time if the processor has popc available.


	Sam